For mixed-signal, roll multiple MEMS dice

For mixed-signal, roll multiple MEMS dice
The increasing use of control systems in vehicles for various functions, from engine monitoring to safety applications, has driven the growth and development of micromachined devices. We have witnessed the transformation of microelectromechanical systems (MEMS) from purely mechanical devices to semiconductor-based transducers. The continuum points toward applied specific integrated sensor (ASIS) devices.

But the perpetual need for smaller, reliable and low-cost components has merged sensor and semiconductor manufacturing, creating a disruptive technology. This article will comment on trends in partitioning between MEMS accelerometer sensing elements and control integrated circuits.

To illustrate two main trends in MEMS manufacturing, we cite the accelerometers used for automotive crash detection. They show one of the partitioning and packaging approaches taken by manufacturers today. The system-partitioning approach can integrate the MEMS and control IC technologies onto a single die or partition the technologies between two dice.

Motorola builds a two-dice accelerometer. The accelerometer sensing element, or g cell, is surface-micromachined in an adapted semiconductor process flow. The original Motorola g cell was a z-axis device, meaning that the plane of acceleration sensitivity is perpendicular to the plane of the chip. Subsequently, lateral (x- and x-y-axis) g cells have been developed. The g-cell wafer substrate is bonded with a cap wafer at the end of the wafer manufacturing flow. The wafer-bonding process uses glass frit bonding technology, providing a sealed environment for the acceleration-sensing elements that protect them during packaging as well as over the lifetime of the operation. This example makes it easy to extract the development directions in partitioning:

• Single integrated die vs. dual die;
• Exposed sensing element vs. protected one by bonded wafers;
• Packaging driven by one vs. the other approach.
• Perhaps the biggest controversy is in the partitioning between the MEMS sensor and the signal-conditioning circuitry.

The single-chip solution may improve signal-to-noise ratio and reduce the number of interconnects. This approach also appears, at first glance, to be a more economical solution. But the multiple-chip system may have a better time-to-market cycle as a result of the flexibility afforded by the separation of the MEMS processing from the circuit fabrication. The sensor and the signal-conditioning circuit can be significantly different in their respective requirements.

Although both single-die and dual-dice ASICs may perform equivalent or very similar functions, monolithic integration is constrained in terms of the extent and level of integration it can offer. In fact, monolithic integration does not necessarily equal sensor smartness of the dual die. The latter can handle system partitioning on multiple chips and thus provide a multiple-chip, system-in-package approach.

Motorola's accelerometer package is based on a separate MEMS sensing element and control IC die. Main blocks of the ASIC, manufactured in the company's SmartMOS7 technology, are labeled, following the signal path from the sensing element to the physical communication and power interface. The control ASIC die performs a number of functions. At a minimum, generically, its role is signal conditioning (amplifying, filtering, offset compensation). Advanced signal conditioning, as in this example, has been made possible by including intelligence at the sensor. Using those methods, signal resolution can be enhanced to the limits of the accuracy of the sensing element and the A/D converter (10-bit resolution in this particular case). In addition, a number of new functions can be added to the traditional signal conditioning.

Integrating intelligence at the sensor component level provides the means for advanced diagnostics of the sensor and transmitter. The diagnostics can be performed automatically or initiated externally, via the user interface. Autodiagnostics can be initiated periodically during operation. One example of this function could be a periodic self-test routine.

These functions would be possible with advanced ASIC technologies (such as SmartMOS7) but would not be possible for some time. The process must also integrate the MEMS sensor. Future development stages will allow the integration of more versatile devices on a single silicon chip. Such is the mixed-signal requirement.

Mixed-signal technology

Mixed-signal processing calls for a combination of analog, digital and power functions on a single silicon chip.

That capability enables integration of microcontroller core, power, linear ICs, communications and nonvolatile memory modules on monolithic chips to meet the needs of highly demanding applications.

Mixed-signal technologies will be vital in creating sensors that are versatile, precise and less expensive than before. Advances in this area will enable new combinations of functionality on a single substrate.

Mixed-signal processes offer 0.4-micron (drawn) geometries, which are compatible with a variety of existing logic libraries. This process has voltage-scalable analog CMOS devices with breakdown voltages ranging from 7.5 volts to 55 V. Lateral and vertical pnp devices can complement high-gain npn devices (beta >100). These processes can include nonvolatile memory E2PROM cells, and many parametric trimming options.

Sensors with digital communication capabilities can eliminate many of the difficulties associated with analog outputs while enhancing the sensor functions. For instance, an integrated A/D converter and state machine lookup table could eliminate the need for an external microcontroller that would ordinarily execute those functions. In addition to signal values, digital communications enable other types of information to be transmitted out by coding them accordingly. Examples of transmitted data include diagnostics results, type of sensor, its ID, manufacturing or application details.

Automotive accelerometers have been in production for over a decade. MEMS-based accelerometers have won acceptance because they provide a means for single-point sensing for airbag crash detection. Many automotive suppliers today have chosen a system-in-package approach: a MEMS transducer chip and CMOS-based ASIC built together in a single package.

Two is often better than one. A two-chip accelerometer avoids the constraints imposed by monolithic integration.

Requirements for such sensors are becoming stricter, not just in terms of reliability and quality but also in the growing number of functions they have to execute. The system-in-package approach reduces the number of external components, But that is possible only through the very advanced dedicated process flows.

Separation of the sensor and control circuitry lets the MEMS technology be separated from the complexity of the ASIC technology. That prevents them from hindering each other's development but fully exploits their capabilities.

Dragan Mladenovic and Dave J. Monk are with Motorola Semiconductor Products Sector (Tempe, Ariz.).