Newton's chips: low-g accelerometer ICs

Mechanical inputs to electrical systems provide devices with an awareness of the world around them. Low-g sensors extend that awareness for a variety of new applications.

By Joshua Israelsohn, Technical Editor -- EDN, October 28, 2004

AT A GLANCE The advent of small, low-g IC accelerometers is stimulating new applications as diverse as security systems and sports and health monitors. The ability to use human movement and gesture as inputs to electronic devices could change the way we interact with consumer electronics, such as portable-entertainment and -communications devices. IC accelerometers are available with analog, PWM, and serial outputs. Consider the format that simplifies your downstream signal-processing task. Low-g sensors by definition are less rigid than high-g devices. Examine the unpowered shock tolerance as a measure of the sensor's mechanical robustness. One of the advantages MEMS accelerometers have over piezoelectric and piezoresistive sensors is the provision for a self-test function. Before choosing an accelerometer, consider the value of in-situ confirmation of basic sensor function in the context of your application's requirements. Sidebars: Measuring motion

Inertial sensors have long been in service measuring shock and vibration in a variety of industrial applications, such as prototype characterization, design verification, and diagnostics for large mechanical systems. Ongoing evolution in sensor, signal-conditioning, manufacturing, and packaging technologies has prompted OEM designers to pioneer new applications.

Before the arrival of the first generation of MEMS (microelectromechanical-systems) devices—sensors that manufacturers fabricate using variations on standard IC-process flows—the most common inertial sensors were accelerometers formed of piezoelectric film. Mechanical deformation of the film causes internal charges to redistribute, resulting in a voltage on the sensor's output (Reference 1).

The piezoelectric accelerometer is capacitive and, like all capacitive transducers, requires a buffer amplifier to drive a useful load. The sensor-amplifier combination exhibits a bandpass response. At midband, the first-order transduction function is v(t)∝kF(t)=kma(t), where v(t) is the output voltage, k is the electromechanical transduction constant in volts per newton, F(t) is the normal component of the force to which the film is subjected, m is the so-called proof mass or inertial mass, and a is the normal component of acceleration.

In typical applications, the piezo element faces a resistive termination. The sensor's capacitance and the amplifier's input impedance combine to set the bandpass's low-frequency corner ω_L=1/R_TC (Figure 1). Three terms can contribute to the spectral location of the high-frequency corner: The buffer amplifier exhibits a dominant pole in its transfer function. The sensor mount and the piezoelectric film exhibit mechanical compliance. Finally, the film's dissipation factor is a loss term that quantifies the extent to which film stress converts to heat at high frequency.

Although applications for piezo accelerometers are numerous, several factors limit them. One factor that is easy to overlook is that bandpass functions—be they derived from linear filters, DSPs, or mechanical systems—have three operating regions, each with its own signal-processing behavior. At midband, the bandpass transduction or transfer function is essentially linear with a signal-independent gain coefficient. At frequencies greater than the high-frequency corner, the transduction or transfer function is that of an integrator—a lowpass filter—which reduces the contribution of high-frequency components by 20 dB per decade per pole. A benefit derives from the lowpass portion of the bandpass function: A dominant pole in or immediately following the buffer amplifier can help reduce the signal chain's noise bandwidth to one only somewhat greater than the measurement bandwidth (Reference 2).

At frequencies less than the low-frequency corner, however, the transduction or transfer function is that of a differentiator. Once the differentiator catches the low-frequency signal and noise components, they can create amplitude disturbances that last many R_TC time constants. These signal components can include such aberrations as turn-on transients; common flicker noise sources, which you can find in both electrical and mechanical systems; and other noise sources exhibiting 1/f- or 1/f²-characteristic spectra (Reference 3). Applications that demand good low-frequency response—some reaching well below 1 Hz—can take many seconds, therefore, to clear disturbances from the highpass portion of bandpass response.

For a given sensor source capacitance, extending the low-frequency response leads to proportionally higher values of R_T. High impedances at the sensor/signal-conditioning interface are more demanding of shielding structures and materials than are lower impedance circuits, which can lead to greater assembly, installation, and maintenance costs.

Piezoelectric films can generate impressively large transient outputs in response to mechanical shock. Sensors with nominal outputs on the order of large fractions of a volt can produce shock responses of many tens of volts. In extreme cases, piezoelectric films have been known to generate sufficient potentials across their terminations to cause arcing. Buffer-amplifier input stages can sustain damage at voltages far less than those required to produce an arc, and input clamping devices add a parasitic capacitance and a leakage-current term that you must consider in your assessment of the buffer's accuracy.

Accelerometer manufacturers have also used piezo-resistive films, which can exhibit spectral responses to dc, do not require resistive termination, and do not generate outputs that are hazardous to signal-conditioning circuits. They typically operate in a bridge configuration and can be subject to drift over time and operating temperature. Depending on the signal-conditioner topology, the bridge can operate with either dc or ac excitation. If the bridge is suitably dissipative, however, self-heating can degrade its accuracy.

Historically, piezo accelerometers have been larger than typical IC sensors, though size comparisons in and of themselves do not fully inform the distinction: Just as whole new applications for components become practical as their price points fall through market thresholds, the same holds true for component size. Unlike price decreases, however, dramatic reductions in device dimensions or mass are relatively rare occurrences, most often requiring significant and near simultaneous advances in device design and fabrication and packaging technologies.

So it has been with IC accelerometers. The first successful commercialized MEMS inertial sensors, primarily from Analog Devices, Freescale Semiconductor, and Robert Bosch, proved that fabricators could build sensors smaller and lighter than their predecessors and do so on conceptually simple—if operationally challenging—variations of standard IC processes. The first MEMS accelerometers were single-axis sensors, most with full-scale ranges of ±35 to ±75g. Though their manufacturers intended these devices for a range of industrial applications, they marketed them most aggressively and most successfully as automotive crash sensors for air-bag-deployment-control systems. Indeed, of the three companies, Bosch is known first and foremost as an automotive supplier.

Compared with current devices, first-generation IC accelerometers provided limited sensitivity, exhibited marginal SNR, depended on expensive and immature processes, and required large and expensive through-hole packages. However, even IC inertial sensors built on early versions of MEMS processes offered substantial advantages over piezoelectric sensors or other motion detectors of the day: Though MEMS inertial sensors are capacitive devices, they can provide a measurement spectrum that extends to 0 Hz. They provide a self-test function that includes the sensing element, which proved an important feature for expensive and safety-critical applications, such as automotive crash sensing. They also form dual-axis sensors with proof mass in a single film.

The 0-Hz response derives from the same fundamental class of circuits that form the basis for isolated dc/dc converters and linear signal isolators, though in this case isolation per se is not a goal. A clock circuit switches between two voltages—say, V_DD and ground for a ratiometric device. The resultant square wave connects to the conductive inertial proof mass etched often from a polysilicon film grown over an oxide layer (Figure 2a). When the fabricator etches away the oxide, anchored extensions, or "tethers," suspend the proof mass over the resultant air gap. Fingers extending from the proof mass interdigitate with fixed fingers, forming parallel differential-capacitor elements (Figure 2b). To first order, as the sensor accelerates along its axis, the proof mass moves relatively in the opposite direction a distance, d, against the tethers' spring constant, k, according to d=ma/k. Also to first order, the differential capacitors model as parallel-plate structures with C∝1/d. Displacing the proof mass increases its capacitance to one fixed finger and reduces its capacitance to the other. The clock's coupling coefficient is proportional to the capacitance and, though the displacement appears in the capacitance's denominator, the sensor's output waveform is reasonably linear for small displacements. A synchronous demodulator recovers the baseband acceleration signal.

Different vendors prefer different signal-conditioning topologies. The figure depicts a demodulator and output-scaling amplifier following an ac amplifier, only one of several possible arrangements. Some designs combine signal-processing functions in a single block, such as a demodulating sensor amplifier, but the essential sensing scheme is similar across different vendors' product lines.

Since the first IC accelerometers that met with market success, sensor makers have been developing more advanced processes, sensor designs, signal-conditioning circuitry, and packaging technologies. The results are low-g sensors that have opened up a large number of applications in several distinct market segments (see sidebar "Measuring motion").

Security applications include motion sensing for briefcases, laptop computers, and other portable equipment, as well as tilt sensing for automobile antitheft systems. Applications in mobile phones and PDAs include hand-gesture sensing for buttonless scrolling and device control, context detection, and gaming. Hard-disk- and CD-drive systems can benefit from active shock or free-fall detection, though the quality of free-fall detection may be limited if excessive rotation accompanies the fall.

Upscale digital cameras and video camcorders can use IC inertial sensors as inputs to image-stabilization processes and have done so since the first IC accelerometers were available. An interesting new application in image processing involves using an accelerometer as a tilt sensor in video projectors for keystone correction.

Helmets equipped with accelerometers and wireless-data links can report on the level of shock sustained during a sporting event and can help coaching staffs limit players' exposure to potentially injurious situations.

In addition to motor-vibration monitoring, industrial applications include automatic gas-shutoff valves that respond to seismic activity. Monitors for small shipping containers can record the time and magnitude of a shock—useful for reconciling claims for damage sustained during transit.

These applications are a few of those that others have found for low-g accelerometers. The list by no means ends there but rather continues as far as your engineering requirements and imagination take you.

The triaxial challenge

Borrowing the traditional spatial designations from mechanics, IC-accelerometer makers have traditionally referred to the sensitive direction in single-axis devices built parallel to the silicon surface as the x-axis, the transverse axis also parallel to the silicon surface as the y-axis, and the axis orthogonal to the silicon surface as the z-axis. Also by convention, the signs follow the right-hand rule: If you point the thumb of your right hand in the positive direction along the x-axis with your palm flat, your fingers will extend in the positive y direction. Curling your fingers 90° points them in the positive z direction.

One way that sensor designers can build dual-axes devices is by forming repeated structures similar to those that Figure 2a depicts along the proof mass's four edges. The resultant devices demonstrate reasonably good axis-to-axis matching of sensitivity, noise, electromechanical spectral response, and robustness. Care in the sensor design can minimize the cross-axis responses—a phenomenon akin to a mechanical nonlinearity through which acceleration along one axis engenders a response in an orthogonal axis. Cross-axis terms differ from electrical crosstalk in that they do not depend on two sensitive axes or two active signal channels. An x-axis accelerometer, for example, may exhibit a small but measurable response to acceleration along the y- and z-axes.

Standard semiconductor processes excel at forming planar structures. One challenge that has kept sensor designers hard at work since the first successful single- and dual-axis accelerometers has been to form z-axis devices with similar sensitivity and robustness. The sensing mechanism for the coplanar axes, x and y, is fully differential. The closest equivalent for z-axis sensing with planar structures is a pseudo-differential structure, which uses the proof mass's capacitance to an electrode formed on the silicon substrate below the sensor and a fixed capacitor designed to match the sense capacitor's nominal value. The exception to this approach, according to Freescale Market Development Engineer Raul Figueroa, is Freescale's differential z-axis sensor used in its uniaxis devices. It takes advantage of a fully differential, three-layer stacked structure with two fixed plates and a movable plate between them.

Though IC-sensor designers have considered single-chip, three-axis devices at least since the mid-1990s, commercially available devices have only recently appeared on the market (Reference 4). STMicroelectronics is one of the few companies making a go at a commercial, monolithic, three-axis accelerometer in the form of the LIS3L02—winner of a 2003 EDN Innovation award. A CDS (correlated-double-sampling) charge amplifier reads the three axes sequentially through a multiplexer. The CDS process, which operates at 200 kHz, cancels offsets and reduces the effects of 1/f-noise sources. Three sample/hold amplifiers acquire the charge amplifier's output at 66 kHz through a demultiplexer and provide the individual axis outputs. A logic-controlled self-test function allows an application to confirm the accelerometer's operation.

The LIS3L02's full-scale range is pin- programmable to either 2 or 6g. ST trims the 2g full-scale range to an accuracy of ±10%. The $4.75 (50,000), triaxis accelerometer has outputs that are ratiometric to the supply, so the nominal 0g level is V_DD/2, also with a tolerance of ±10%. The typical noise density is 50 µg/; the typical sensor resonant frequency is 4 kHz for the x- and y- axes and 2.5 kHz for the z-axis.

Available in SO-24 or 7×7-mm QFN-44 packages, the LIS3L02 typically draws 850 µA from 3.3V supplies. The operating-voltage range is 2.4 to 5.25V. In addition to the version with analog outputs, STMicroelectronics provides a $5 (50,000) version of the LIS3L02 with PWM outputs and a $5.50 (50,000) version with I²C- and SPI-compatible serial outputs. The serial version integrates three delta-sigma converters and user-programmable decimation filters that provide a signal bandwidth variable over 280 Hz to 4.48 kHz.

Since STMicroelectronics announced its triaxial accelerometer, two companies have also announced triaxial devices, albeit with scant technical details available at press time. Oki Semiconductor and Hitachi Metals describe ±3g devices with nearly identical drawings of their sensor structures. Hitachi Metals describes its sensor as a piezoresistive device with a bandwidth of "several hundred hertz." Oki Semiconductor puts the bandwidth of its device at 200 Hz. Neither company has yet provided limit specs, tolerances, or noise figures.

In the plane; out of the plane

More common are the single and dual in-plane (x and x,y) sensors and single out-of-plane (z) devices. Single-axis z devices and triaxis devices were slow to emerge for many of the same reasons: Structures for sensing motion orthogonal to the silicon plane are difficult to come by and, with the exception of the aforementioned Freescale design, tend to depend on a balance between dissimilar structures.

Z-axis devices require reasonably compliant tethers to make a useful signal. In-plane sensors, on the other hand, benefit from tethers that are stiff in the z-axis, which help prevent the proof mass from landing on the substrate in the event of a large-amplitude vertical shock. Surface effects well-known to MEMS process engineers can cause small thin films, such as those that form proof masses, to adhere to smooth surfaces even in the presence of the return forces the stretched tethers supply. The phenomenon, stiction, is due to a combination of sources including capillary, electrostatic, and van der Waals forces (Reference 5). Left unchecked, stiction can render a MEMS accelerometer permanently nonfunctional—a problem that has given rise to various treatments ranging from geometric features in the sensor design to thin coatings that alter the sensor's and substrate's surface chemistry. The antistiction coatings are most often proprietary. With an increasing number of foundry fabs adding MEMS modules, open-market antistiction coatings are emerging from companies such as Surmet (www.surmet.com), which can save process-development engineers from reinventing that particular wheel. Examine the maximum unpowered shock specification for a measure of sensor robustness in the face of potential stiction mechanisms.

Other sensing methods, such as the air-mass sensor technology Memsic uses, don't suffer from stiction because the proof mass is not a solid. Unfortunately, such sensing methods do not provide for z-axis sensitivity.

Freescale's z-axis MEMS sensors currently include 1.5, 2.5, 5, and 8g devices. The 1.5g MMA1260D accelerometer exemplifies the family. The 1260's integrated signal conditioner provides a two-pole, lowpass Bessel filter; temperature compensation; and a self-test function. The bandwidth is factory-trimmed to 50 Hz nominal. The noise spectral density is 500 μg/. Nonlinearity is limited to 1%, which is about a third of the z-axis nonlinearity of the only triaxial accelerometer to offer the spec at all.

The nonlinearity measure could be of importance in measurement applications but is generally not in applications such as gesture sensing. For example, a wireless phone could use a z-axis accelerometer to detect when a user places the phone face-down on a horizontal surface, a gesture that could signify the user's hanging up the phone without pressing a key. In such a case, the gesture sensing system needn't make precise linear measurements at all angles but merely detect the earth's gravity.

The $5.68 (1000) MMA1260D requires only three external components: a bypass capacitor and a resistor and capacitor to form a lowpass filter to remove clock artifacts on the output. The accelerometer operates on 5V nominal supplies and draws a maximum of 3.2 mA. Zero-g offset, full-scale span, and filter cutoff are factory-set and require no external trims.

Unlike most of its competitors, Freescale has always built its MEMS accelerometers as separate sensor and signal-conditioning chips and packaged them together in standard IC form factors. There are pros and cons to both the monolithic and so-called compound-monolithic approaches. The monolithic structures, favored by Freescale's competitors, such as Analog Devices, benefit from smaller strays along the path that connects the sensor's small output capacitance to the signal conditioner's input. Additionally, assembly costs are less for monolithic devices than for multichip assemblies. In favor of dual-chip approach, the two chips can take advantage of different process flows—each optimized to the device types it must fabricate. With significant market opportunities evolving for the smallest sensors, compound monolithic structures have evolved from side-by-side assemblies to bonded-wafer and stacked-chip approaches (Figure 3).

Sensing attitude

Some low-g applications can use sensors with bandpass responses, but tilt detection is an example that demands a measurement bandwidth that extends to 0 Hz. You can align a linear accelerometer in two ways to make a tilt measurement (Reference 6). Aligning the accelerometer's sensitive axis parallel to the earth's radius, the response to a tilt angle, θ, is V(θ)=V_O+S_1G(1–cos(θ))g, where V_O is the accelerometer's 0g output (a nonzero value for unipolar power supplies) and S_1G is the accelerometer's sensitivity in volts/g. Tilt measurements, however, most often concern small angles and, with such an alignment, the system must resolve a small difference in large numbers.

The alternative is to align the accelerometer orthogonally to the earth's radius. In this case, the response to a tilt angle is V(θ)_O1G=V_O+V_1Gsin(θ)g. Though the electrical bias, V_O, is still present, the 1g measurement bias is not. In some cases, such as automotive antitheft systems that need to detect jacking or towing, the orthogonal alignment can somewhat desensitize the system to false inputs, such as road vibration due to passing heavy vehicles or other sources of mechanical noise.

Most tilt-measurement applications are limited to an angular range of about ±10°, which corresponds to a gravitational measurement of ±0.17g. Recalling that for small angles expressed in radians, sin(θ)≈θ; a resolution of ±1° (±0.01745 radians) requires a measurement resolution of about ±17 mg. First-generation sensors for full-scale measurements in the tens of gs could not make these kinds of measurements except by averaging over long periods. For example, one of the first commercially successful IC accelerometers, the Analog Devices ADXL50, offered a maximum noise density of 12 mg/ and a sensing range of ±50g. Ignoring all other noise sources including 1/f terms, the sensors self-noise would give a 3-dB SNR in a 1-Hz noise bandwidth. Clearly, Analog Devices never intended the ADXL50 as a tilt sensor, but the comparison with modern low-g devices indicates how far sensor design has come in a short time.

Better suited to the task, the ADXL103 is a ±1.7g sensor with a noise floor of 110 μg/. If, for example, you want to make a tilt measurement with a minimum SNR of 20 dB, the ADXL103 can resolve 1° in a 250-Hz noise bandwidth. In the same 1-Hz bandwidth as the previous example, the ADXL103 yields a slightly better-than-40-dB improvement in SNR in the face of a 34-dB increase in sensitivity. Despite its high sensitivity and low noise, the ADXL103 can survive an unpowered shock of 3500g.

The $7.75 (1000) accelerometer requires only two external components: a power supply bypass capacitor and an output filter capacitor. An integrated nominal 32-kΩ resistor completes the filter. Your selection of filter capacitor sets the sensor's bandwidth with two caveats: The integrated resistor's ±25% tolerance combines with the external capacitor's tolerance to determine the total variability in the filter's pole location. The accelerometer requires a minimum filter capacitance of 2 nF, corresponding to a nominal –3-dB bandwidth of just less than 2.5 kHz. If your application does not demand that great a bandwidth, you can reduce the effects of flatband noise by further reducing the signal bandwidth. In all cases, ensure that the device cannot pass significant signal (or noise) energy at frequencies greater than the Nyquist rate for your application's ADC.

Both the ADXL103 and its dual-axis sibling, the $12 (1000) ADXL203, are available in 5×5-mm CLCC-8 packages. The accelerometers draw a maximum 1.1 mA from nominal 5V supplies.

Freescale's nearest equivalents to the ADXL203 are the MMA626x family of dual-axes ±1.5g accelerometers. Four devices, number 6260 through 6263, differ primarily by bandwidth: 50, 300, 150, and 900 Hz, respectively, which a one-pole, on-chip, switched-capacitor filter determines. The 60 and 61 have nominal noise spectral densities of 300 μg/; the 62 and 63 offer 200 μg/. The maximum supply currents are similarly paired: 1.5 and 3 mA from 3.3V nominal supplies. The maximum guaranteed shock tolerance is 2000g. Freescale provides the $3.60 (1000) MMA626x family in QFN-16 packages.

Most low-g accelerometers have a package alignment tolerance of 1°. Your soldering process introduces another angular tolerance as does the process of mounting the pc board in your application (Reference 7). Though each package type brings its own soldering requirements, some concepts apply more universally.

If yours is a portable application, then the attitude of your device may also be a variable. Users may turn cell phones and PDAs to accommodate the user's position, lighting conditions, or other environmental variables. In such cases, a measure of absolute angle with respect to an inertial reference is less useful than relative measurements a device may periodically make during its use. If you are going to use human motion as a system input in your application, think carefully about the gestures and positions that are likely in everyday use and, if appropriate, what behavior or motion signature may signify a discontinuation of use.

You can reach Technical Editor Joshua Israelsohn at 1-617-558-4427, fax 1-617-558-4470, e-mail jisraelsohn@edn.com.

 

 

MODULAR ACCELEROMETERS AND VALUE-ADDED RESELLERS
CFX Technologies www.cfxtech.com	Crossbow www.xbow.com	EDC www.smartsensor.com
i-for-T GmbH www.i-for-t.com	MicroStrain www.microstrain.com	Sequoia Automation www.sequoiaonline.com
Silicon Designs www.silicondesigns.com	VTI Technologies www.vti.fi

References

"Piezo film sensors technical manual," Measurement Specialties, 1999. Israelsohn, Joshua, "Noise 102," EDN, March 18, 2004, pg 46. Israelsohn, Joshua, "Noise 101," EDN, Jan 8, 2004, pg 41. Boser, Bernhard, "3-axis accelerometer with differential sense electronics," Berkeley Sensor and Actuator Center, University of California—Berkeley, 1996. Gunda , Nilesh, et al, "Anti-stiction coatings for high reliability mems, " Surmet Corp. Weinberg, Harvey and Christophe Lemaire, "Using the ADXL202 accelerometer as a multifunction sensor (tilt, vibration, and shock) in car alarms," Technical Note, Analog Devices, 1998. Geitner, Hubert, "Considerations for soldering accelerometers in LCC-8 packages onto printed circuit boards," Application Note AN-652, Analog Devices, 2003.

Acknowledgments

Thanks to Mike Markowitz at STMicroelectronics, Raul Figueroa at Freescale Semiconductor, and Christophe Lemaire at Analog Devices for their contributions.