5. SENSOR-CIRCUIT INTEGRATION AND SYSTEM PARTITIONING
(Kensall D. Wise)

Throughout their history, sensors have traditionally been viewed as components to be purchased as needed and added to existing systems to complete their interface with the nonelectronic world. This view persists in many areas of industry, especially in the United States, and yet there is a growing recognition that these devices should in fact be viewed as system elements and be developed as an integral part of the overall system design. As components, sensors are rather limited in function by the fixed systems that represent their market, and there is limited ability to do more than replace earlier components. Thus, the interfaces remain 5 V analog signal lines connected by point-to-point wiring, and auxiliary functions are difficult to introduce. Such arrangements do not take full advantage of the capabilities of silicon as a transducer substrate, and do little to spark the progress that the sensor/actuator field should represent. On the other hand, the ability to build additional functions into the transducer module, such as self-testing, autocalibration, digital compensation, and digital bus compatibility, would both improve system reliability and decrease system cost.

Many major companies in the United States are today in the process of developing mixed sensor/actuator systems (MEMS) that merge electronics with transducers to produce microinstrumentation systems in a single module and, in some cases, on a single chip (Wise and Najafi 1991). This includes major efforts at most of the automotive companies, in the environmental controls industry (HVAC), and in the design of industrial process controls. SEMATECH, for example, is working to define a sensor bus for intratool use in semiconductor process equipment. Such buses will replace point-to-point wiring between dozens of sensors and actuators, and the host controller, with bus communication. It would appear that in applications where distributed sensing and actuation are required and where hierarchical control is used, such microinstruments will become commonplace within a decade, at least in simple forms. Figure D.6 shows a diagram of such a system (Najafi and Wise 1990). Because of their suitability for miniaturization, these devices will likely concentrate initially on sensing, with actuation used at first as a means for providing functions like self-testing. Exceptions are likely to be the use of microvalves for the implementation of flow control or in instruments such as sampling systems for use in medical diagnostics or in miniature gas chromatography systems. Inertial guidance systems will make extensive use of microactuators to provide drive to the sensors, as will other forms of resonant sensors.

From a system point of view, the level of integration used in these microsystems will be decided primarily by cost and reliability issues. It is interesting that in industries where volumes are high and reliability is paramount, monolithic implementations are being pursued. In situations where expected volumes are lower, hybrid approaches are being explored, as might be expected. Processes for merging transducers and circuits are being pursued based on bulk micromachining (Wise and Najafi 1991b; Ji and Wise 1992), surface micromachining (Mastrangelo and Muller 1991; Yun, Howe, and Gray 1992; Nguyen and Howe 1992), and combinations of the two (Kong, Orr, and Wise 1993). Figure D.7 shows typical process flows. In either case, the goal is to modify the circuit portion of the process as little as possible in order to make the overall process as directly compatible as possible with those in commercial foundries. This is important to allow adoption of the technologies by small to medium size companies that may not have their own fabrication facilities, and is also important for lower-volume applications in which the development costs for a full-custom process might be more difficult to recover. For bulk micromachined processes, some diffusions must typically be performed at the front-end of the process flow to embed appropriate etch-stops in the material. The circuit process then continues without interruption until final metal. This must either be a material that will not be attacked in the silicon etchant used or it must be protected using dielectric overlays. In the latter case, special inlays at the bonding pads are still required. The micromachining etch must be performed as the last step in the process, where it can also serve for die separation if desired. For surface micromachined structures, the transducers are formed using post-processing steps, after circuit formation. Here, special metal may still be required to permit high-temperature annealing of polysilicon microstructures (Yun, Howe, and Gray 1992). In this case, a high-quality dielectric overcoat (e.g., with silicon nitride) of the entire chip is required to shield it from the final sacrificial etch. This release should be performed before die separation, but in this case the delicate microstructures must be protected during the die separation procedure used. This is not a simple challenge, but has been successfully implemented commercially (Payne and Dinsmore 1991).


Figure D.6. General organization of a distributed sensing system. Each node here is a monolithic or hybrid module comprising a microelectromechanical system of sensors, actuators, interface circuitry, and a microcontroller/microprocessor chip.


Figure D.7. Process flows for merging bulk- and surface-micromachined transducers with on-chip circuitry.

Figure D.8 shows an uncooled monolithic thermal line imager implemented recently using on-chip CMOS circuitry and bulk-micromachining. Each of the thirty-two stress-relieved dielectric windows supports forty polysilicon-metal thermocouples that convert the temperature rise on the window induced by incident radiation into an electrical output signal (Baer et al. 1991). The windows measure 400 mm x 800 mm with a responsivity of 60 to 80 V/W, a time constant of about 5 msec, and a remote temperature resolution of about 0.2C. On-chip circuitry multiplexes the output signals, measures the ambient temperature, and can provide self-testing. An alternative to such back-etched structures are devices that are based on front-side undercut devices (Baltes and Moser 1993). These are being pursued in both the United States and Europe because of their compatibility with standard foundry processing up to the final front-side etch, but have been used more for thermal devices than for mechanical structures. Figure D.9 shows a thermally-based absolute pressure sensor complete with MOS detection circuitry recently developed at the University of California at Berkeley (Mastrangelo and Muller 1991). This device uses a heated micromachined polysilicon beam to measure the thermal conductivity of the surrounding gas and hence its pressure. Another example of this technology is found in a microelectromechanical filter chip, also developed at Berkeley, which contains a variety of electrostatically-driven mechanical elements together with integrated electronics (Nguyen and Howe 1992).

One of the higher levels of circuit integration on a micromachined structure is shown in Figure D.10. Multichannel silicon probes are under development for the stimulation of biological neural networks with high precision, both spatially and in terms of charge delivery to the tissue (Tanghe and Wise 1992). The probe structure is defined using a deep boron diffusion in a single-sided bulk process and incorporates CMOS circuitry to provide per-channel 8-bit current control over sixteen electrode sites using only five external leads. The circuitry accepts serial input data at 4 MHz to select the site and current level desired and dissipates only 80 mW from 5 V supplies when idle. The currents generated cover a biphasic range of 254 mA with a resolution of 2 mA. The circuitry permits the IrO sites to be activated by voltametry from off-chip, provides per-channel pulse timeout to prevent accidental overstimulation, signals the external world in the event of certain trouble conditions, and is implemented using about 7,000 transistors in a circuit area of 11 mm(2) in 3 mm features. Solid-state sensors have evolved over the past two decades from devices having one-way analog outputs through digitally-addressed analog-output modules to fully integrated sensor/actuator systems with onboard microcomputers or microcontrollers. Although relatively high levels of monolithic integration are being pursued in connection with the last devices, it seems likely that the levels of integration on monolithic MEMS chips will remain relatively low prior to 1995, reaching levels above 10,000 transistors only after that date. The integration of full microcontrollers on such chips seems likely for some applications by the year 2000, but even at that time, most MEMS applications will likely remain hybrid, employing standard commercial processors together with front-end MEMS chips containing the transducers along with a modest amount of electronics. Figure D.11 shows such a configuration. The identification of appropriate high-volume targets for commercialization remains a challenge and a concern in pacing industrial developments.


Figure D.8. Top view and cross-section of a 32-element thermal line imager formed by series-connected thermocouple arrays supported on micromachined dielectric windows. A top view of a portion of the imager is shown below with the windows back-lighted. The windows measure 400 mm x 800 mm in size.


Figure D.9. Top view of an integrated thermally-based absolute pressure sensor with MOS readout electronics (Mastrangelo and Wise 1992).


Figure D.10. Drawing and top view of a micromachined neural stimulating probe containing 16 8-bit CMOS digital-to-analog converters and associated control circuitry.


Figure D.11. Structure for emerging high-end microsystems, employing a front-end chip containing transducers and interface electronics along with a microcontroller chip to give the module flexibility and stored-program intelligence.

Bus standards will certainly be adopted before the end of this decade. These will probably be specific to particular industries. Coordinated efforts such as the one at SEMATECH could unify standards within a particular industry, but commonality among different industries appears unlikely. Even in the automotive industry, several different bus standards are still being promoted, and the requirements there are viewed as different from those in the industrial process control area. In automotive, each node must be able to communicate with any other node, whereas in industrial controls it is probably sufficient if each can communicate only with the host controller, and then primarily in response to commands received from it. Greater interface commonality is likely to be achieved through the use of programmable interfaces, and perhaps through overdesign for some applications.

One of the primary reasons for embedding a microcontroller in the sensing node is to improve the accuracy of sensor/actuator calibration procedures. Historically, all sensors have needed compensation for offset, slope errors, and output temperature sensitivity. These trims have normally been implemented using laser trimmed resistor networks. Such compensation is typically adequate only if the required compensation is linear and hence is viable only over a restricted dynamic range. The use of digital compensation using lookup tables or polynomial coefficients stored in EPROM can follow very nonlinear sensor outputs. The first digitally-compensated devices are now beginning to appear, and have produced roughly order-of-magnitude improvements in accuracy (Wise and Najafi 1991a). In theory, such an approach should allow the system to utilize the full performance available from the transducer, limited only by the device stability over time. As in other areas of microelectronics, sophisticated procedures once reserved only for high-end devices are likely to pervade even low-end devices within the next decade. The area of testing and compensation has received very little attention in the literature in the United States and probably globally. But it is of great importance to coming systems, and needs to be addressed more vigorously.

During the past year, several efforts to make MEMS available through foundry services have been undertaken. The most notable efforts are at MOSIS and at MCNC. The goal of these efforts is to make MEMS much more widely available, both to university researchers and to industrial organizations seeking a rapid prototyping service. As part of these efforts, it is likely that a few standard processes will be employed. As in the microelectronics industry, larger companies will optimize their internal processes to meet their own particular needs. Even in such companies, however, only a few different processes will be supported, and they are likely to differ from company to company.

Thus, in the United States there appears to be a pronounced trend toward merging transducers and circuits monolithically, and toward merging these chips, employing modest levels of integration, with commercial microcontrollers to form microinstruments (Wise 1993). Most of these efforts are still in the research stage, with commercial products expected during the latter half of the decade. After that time, a continuous series of enhancements are expected.


Published: September 1994; WTEC Hyper-Librarian