A Practical Taxonomy for Microsystem Applications
(Extended Abstract)

J.C. Boudreaux
NIST/Advanced Technology Program

1. Introduction

A microsystem is “­a collection of highly integrated devices that contains transducers along with appropriate interface circuitry and is capable of perform­ing multiple tasks autonomously as well as responding intelligently to various commands from a host sys­tem.” /5/  Compared to traditional devices, microsystems can be fabricated from small amounts of raw materials, use much less energy, and generate smaller volumes of waste when disposed of. But the commercialization of microsystems technology has been slower than originally forecast.  Many universities, national laboratories, and for-profit companies have been successful in developing specific microdevices for specific applications, but with a few notable exceptions, it has proven to be difficult to turn these devices into commercially viable products.

There are substantial technical barri­ers to be overcome.  These barriers arise in the first instance because of the novelty of the technology, which may seem to be yet another case of resistance to inno­vation, but in fact is a reaction which ­is an accurate reflection of the inherent complexity of the behavior of mechanical systems in micrometer and submicrometer domains. Second, from the perspec­tive of the technology provider, the main technical barrier is to understand the engi­neering requirements of the application system, and, more narrowly, to understand them as defining technical constraints on acceptable microsystems designs. This abstract is an attempt to provide an initial sketch of a framework for the discussion and analysis of this set of issues. 

2. Classifying Microsystems

Many different classificatory principles have been proposed to separate the family of microsystems into meaningful groups. For example, the most frequently used principle distinguishes microsystems based on their method of fabrication.  That is, devices fabricated by Si-based semiconductor (IC) technologies, including both surface and bulk micromachining processes (photolithography, thin-film deposition, implantation, etching, and planarization) are distinguished from devices fabricated by non-Si-based technologies, including LIGA and precision CNC machining. A second approach distinguishes microsystems in terms of their mode of integration: monolith­ic systems in which the components are all co-located on a single chip are distinguished from multi-chip modules whose components are electri­cally connected.  Other interconnection modes have also been investigated, including wireless radio-fre­quency (RF) links which can potentially be used to both supply power to sensors and to read out their information, either on demand or on a preset schedule /5, 7, 10/.

While both of these approaches have been useful, a different approach seems more directly relevant to the theme of this paper, specifically, a taxonomy that takes seriously (1) the fact that microsystems are engineered systems, and also (2) that microsystems must have proximate computational capability.  Let’s summarize the implications of each of these two observations.

First, every system is defined by separating it from the rest of the universe by means of a physical or conceptual boundary.  Everything outside the boundary is part of the system’s environment.  Engineered systems are assumed to be open, that is, they receive signals from and emit signals to their environments. These boundary-crossing events are of two kinds:  (1) signals which are produced and consumed by external systems or system components, called inputs or outputs; and (2) signals whose sources and sinks are not localized to other systems or system components.  Signals of the first kind are mediated by a two-sided I/O link, one side of which is owned by a write process and the other by a read process.  Signals of the second kind are unmediated events whose presence is indicated by unwanted disturbances to I/O signals and also to the system’s operational state. For example, engineered systems are affected (disturbed) both by the flow of thermal energy from the environment and the radiation of thermal energy into the environment. /1, 11/  The behavior of a system is defined in terms of behaviors of its components and the rules which govern their interconnection.  In practical applications, the behavior of every system as a whole and each of its components is described by a transfer function which, given a time-sequence of inputs, and possibly information about the current state of the component, produces a time-sequence of outputs (and possibly a new state) after a specified time lag. /1, 2, 3/

Second, a microsystems’s computational capability is provided by processors, that is, microdevices which (1) acquire and store inputs, (2) apply algorithms to the stored data to generate output values, and then (3) transmit the computed outputs. Because processors are computational devices, the inputs and outputs can have several different forms.  Valid inputs and outputs will obviously include digitalized versions of sensor (input) and actuator (output) signals.  Computational elements are almost always digital electronic systems, but other approaches are technically possible. For example, some interesting work has been done using analog electronics.  Processors may be designed to accept other I/O tasks. For example, processors may accept command inputs whose purpose is to modify or change the  algorithm or to update the contents of the processor’s store (memory).  In general, the effect of command inputs is usually described as a change of state.  Processors which have been designed to accept inputs that affect the algorithms to be executed and not merely the stored data are said to be programmable.  Processors may also be assigned the task of preparing output reports for external (open loop) consumption. /9, 11, 12 /

Using this framework, microdevices may be classified by their input/output mode of operation. Two microdevice families may be distinguished.  First, sensors are microdevices that react to environmental stimuli to produce a time-dependent range of values.  Once properly calibrated, there is a functional connection between the response values and the corresponding physical state of the stimuli.  Each sensor has sensory modality, that is, a range of stimuli to which it is able to respond.  Modalities of interest include pressure, accel­eration, force, angular acceleration, torque, vibration, stress, strain, fluid flow, thermal properties, and ambient radiation. /3, 4, 8/. Second, actuators are microdevices which act on and modify the environment.  Actuators may act directly to cause an effect.  For example, the airbag trigger is designed to sense acceleration and then to initiate an ignition sequence if the sensed acceleration is greater than a preset threshold value.  But actuators can also be designed to act indirectly, for example,  by producing a signal in response to a sensed condition which modifies the feedback signal from a pre-existent host controller. /1, 2, 3, 5, 7/

Microsystems in which the sensory modes alone are present are called monitors and are defined by an operational cycle which is a sense-compute-report loop. Some condition of the host system is sensed, relatively shallow computations are then performed on the sensed data, and periodically, or perhaps aperiodically on the detection of abnormal situations, status reports are passed to a host supervisory agent.  The broad classification of monitors can be refined along several dimensions, for example, increasing the number of sensors often introduces significant increases in complexity, even if the sensor modality is the same in each case. /4/ Microsystems which also contain actuation components are called controllers and are defined by an operational cycle which is a sense-compute-act loop./1, 5, 9/  That is, they augment the monitor functions with the capacity to generate control signals to modify the operational behavior of the host.  Since they produce changes in the external world, they must be validated against hard real-time standards: given any feasible inputs, it must not only be demonstrated that the corresponding outputs are mathematically correct, but that these outputs can be reliably generated by a specified time.  In effect, the art of hard real-time system design is to successfully exploit the difference in payoff between correct but late outputs and outputs which are (modestly) incorrect but timely. /1, 9/

Microsystems can also be classified by their environments.  Though difficult to define with technical precision, this classification allows us to distinguish stand-alone devices, such as medical devices (diagnostic chip sets, implantable monitoring devices),  biotechnology devices (DNA diagnostics), and chemical devices (multifunction “lab-on-a-chip” detectors, combinatorial chemistry panels) from embedded devices that are designed to be components of larger host systems. The former class of devices projects into and  interacts with a “natural” environment, whereas devices of the latter class project into what is in part an “artificial” or “engineered” environment.  That is, relevant parts of the environment of embedded microsystems have themselves been built to engineering specifications.  Thus, there is at least the practical possibility that a host and its “symbiotic” microsystems can co-evolve.

3. Toward a General Theory

This approach to the classification of microsystems can be formulated in a compact and suggestive style by developing a small grammar.  Grammars are mathematical constructions in which syntactic rules are used to derive a collection of “well-formed” taxonomic expressions by successive symbol replacements.                

I           top   ::=   microsystem link environment

II           microsystem   ::=   monitor | controller

III          monitor   ::=   sensor link processor

IV         controller   ::=   monitor link actuator

Given the remarks in the previous section, the interpretation of the rules should be clear.  Thus, for example, the first says that the expression top may be replaced by the expression microsystem link environment, which means in effect that microsystems are linked to an environment; and the second says that the expression microsystem may be replaced wherever it occurs by either monitor or controller (but not both), which means that every microsystems is either a monitor or a controller.  One can see at a glance that the following derivation is correct:

1. top
2. microsystem link environment
3. monitor link environment
4. sensor link processor link environment

At this point the derivation must stop because the first four rules do not show us how to rewrite any of the symbols in expression (4).  If at step (2) we had chosen the second alternative of rule II, then the derivation would have proceeded as follows:

3. controller link environment
4. monitor link actuator link environment
5. sensor link processor link actuator link environment

Again the derivation stops.  Since a single microsystem can contain many sensors, and for that matter many actuators and processors, it would be appropriate to add rules which permit this to happen:

V          sensor   :=   sensor sensor

VI         actuator   ::=   actuator actuator

VII        processor   ::=   processor link processor

That is, any place in a syntactically well-formed taxonomic expression in which the symbol sensor (actuator) occurs remains well-formed if another occurrence of sensor (actuator) is inserted next to it.  Any place in a well-formed expression in which the symbol processor is rewritten by the expression  processor link processor remains well-formed.  Rule VII reflects the requirement that it must be possible to link multiple processors to each other. With these new rules in place, the derivation from step (5) may be continued as follows: 

6. sensor sensor link processor link actuator link environment
7. sensor sensor link processor link  actuator actuator link environment
8. sensor sensor link processor link processor link  actuator actuator link environment

and so on.  In the initial system only finitely many well-formed expressions could be derived.  In the extended system infinitely many well-formed expressions can be derived, all of which have an exceedingly regular syntactic structure. For example, it is easy to see that expression (8) can be unambiguously recovered from much briefer version

9. [S(2)P(2)A(2)] link environment

and that with suitable choices of parameters, (9) can be generalized to

N.         [S(I)P(j)A(k)] link environment 
Every expression that can be derived in the extended system can be put in the form of (N).  It seems likely that every imaginable microsystem is an instance of one and (maybe) only one of these expressions, i.e., simply count the sensors, actuators, and processors, and fill in the parameters of (N) accordingly. 

However, given any specific value of (N), the fact that two microsystems are instances of it says very little about their structural affinity.  To handle this, we’ll need to understand the syntactic role of the symbol link.  Traditional grammars are usually designed to handle languages or language-like formal systems in which the instances of derived expressions are linear sequences of words.  Each word is paired with a nonterminal syntactic category of the grammar (e.g., noun, verb, adjective), which play very nearly the same roles as the nonterminals of our rule system. The words are then collected into a compilation called a lexicon.  To generate grammatical (syntactically well-formed) sentences is a two-stage process: derive a syntactic expression, and then replace each nonterminal with any word paired with it in the lexicon. The fact that both the syntactic expression and the sentence are linear structures makes this an easy task, i.e., words are two-handed constructs and link with each other like beads on a string. But microsystem “words” are microdevices, i.e., sensors, actuators, and processors, and thus may be many-handed constructs.  The same polarity obtains, i.e., every hand is either a left hand (input) or a right hand (output), but  “words” may have many left and right hands.  Thus, every “word” must be paired not only with a syntactic category (e.g., sensor, actuator, processor) but also with a description of its handedness.  For example, if A is a sensor then its handedness can be expressed by an expression of the form [L (1), R(1,2)] which says that A has one left hand, carrying the label “L1", and two right hands, carrying the labels “R1" and “R2", which by the polarity convention says that L1 is the source of an input signal to A and R1 and R2 are sources of output signals from A.  A lexical entry to record this might look like this:

            define  A =  sensor[L(1), R(1,2)]

Similar entries could be used to record the handedness of the two other classes of microdevices.

To generate microsystem “sentences” is a three-stage process, the first two of which are exactly the same as above.  The third stage links the sequence of “words” derived at stage two by selecting a suitable pairing of hands and then linking them based on that pairing.  First, we build the set of all possible pairings. Since every “sentence” has finitely many “words”, there is a unique maximal set of all ordered pairs whose first component is a left hand and whose second component is a right hand.  This maximal set has two useful properties: (1) only hands of opposite polarity are linked, and (2) each “word” is linked with at least one other “word”.  Let’s stipulate that suitable pairings are exactly the subsets of the maximal set that also satisfy (1) and (2).  The mathematical effect of this construction is to transform  “sentence” from its core representation as a linear sequence to a representation based on (finite) directed graphs, a more complicated family of structures.

I suppose that any linking which begins with a suitable pairing should be considered (marginally) well-formed, but many of them make no sense at all.  For example, the maximal set is a suitable pairing and would have the effect of linking every left hand to every right hand!   Some more restrictive properties have to be added, including (3) every left hand of a sensor must link to the environment and every right hand to a processor; and (4) every right hand of an actuator must link to the environment and every left hand to a processor.

The next step would be to begin the construction of a “semantic” theory for the microsystem “words” themselves.  Such a theory must support the compositionality principles, which are (1) the behavior of every microdevice and each of its components is described by a transfer function which, given a time-sequence of inputs (left-hand values), and possibly information about its current state, produces a time-sequence of outputs (right-hand values), and possibly a new state, after a specified time lag; and (2) the behavior of a microdevice is defined in terms of behaviors of its components and the rules which govern their interconnection.  Theories of this sort have been thoroughly elaborated in the domain of electronic microdevices and circuits.  Even though a similar effort in the domain of general microsystems would be a very much more difficult undertaking, it is a task that the MST community must now be prepared to confront head on.

4. Classifying MST Insertion Points

The proposed  taxonomy obviously needs to be deepened and refined, a project to be undertaken in subsequent efforts.  But  for now it might be more useful to at least sketch a possible classificatory structure of microsystem insertion points.  Henceforth, the discussion will focus on embedded applications that involve high-volume, cost-sensitive mechanical systems. Hosts include both end products (automobiles, household appliances, industrial machine tools) and also components of end products.  As a group, these applications present harsh environmental conditions for microsystems, including high temperatures (greater than 200 C), high g forces, high electromagnetic radiation, high static pressures, and chemically corrosive or mechanically abrasive conditions.  In the context of microsystem technologies, innovation amounts to a strategic review of the product architecture in order to locate the components that would most benefit from technological development. Such components will be called insertion points. The critical issue for microsystem technologists may be seen as an ecological one: how best to traverse the “landscape” of the host to effectively and efficiently identify high-yield, company-building insertion points.

The commercial viability of such devices is based on their ability to add value to host applications, and the innovation pathway for MST technologists is a matter of identifying feasible, and profitable, device insertion points within suitable hosts.

First, an often-overlooked class of insertion points is provided by the host firm’s industrial equipment.  The technical target in this case is to enhance both the flexibility of the capital equipment and the span of control and scope of action of the workforce. In addition to performance enhancements, a plausible business case for “process” MST insertions could also be based built on a cost-avoidance story. Unlike the cost structure for consumer products, these plant-related process equipment insertion points share the same harsh environment as do the target groups, but are needed in much lower quantities and will support a higher price structure. As is the case for all capital equipment, the cost is distributed over all the large number of products manufactured.   It would also be important to be able to establish that the proposed microsystem, when installed, would encourage the avoidance of maintenance and other recurring costs of operation.

A second class of insertion points will add value directly to the products.  This requires the MST technologist to have an informed sense of the context within which the firm and its competitors are operating. When a firm decides to enter the market for a specific family of products, it needs to learn how to participate in a value network, which is  “the context within which a firm identifies and responds to customers’ needs, solves problems, procures input, reacts to competitors, and strives for profits.” /6/.  Application engineers and MST technologists are working collaboratively: the former owning the insertion points, the latter owning the expertise in microsystem technologies.  A winning play is to discover and develop a hitherto unexploited, or underexploited, product attribute that can then be boosted to otherwise unreachable performance levels by the development of an innovative Microsystem. With appropriate intellectual property protections in place, the microsystem can serve as a technological barrier against the firm’s competitors.

A third class of MST insertion points are those that arise as a response to specific governmental regulations and mandates. Governments do on occasion mandate a certain performances for all or some of the products of an industry sector, e.g., imposing product safety regulations, such as airbags or antilocking breaking systems, or environmental and fleet-wide fuel consumption standards. Since these burdens will fall on all firms more or less equally and the penalties for noncompliance are stiff, these insertion points are attractive to MST technologies who are able to demonstrate the capacity to manufacture microsystems in the necessary quantities, at the right quality, and who can also make a cost avoidance case.

5. Discussion

To a first order approximation, every current microsystems device is a highly customized product. To tap the real potential of this technology, it is imperative that it outgrow the customization stage. To ensure wide accessibility and commercialization, microsystem technology needs to evolve towards a foundry model in which technical resources are principally committed to design and production of actual devices rather than into the speculative development of a myriad of additional custom processes. The foundry model is based on a focused set of processes, which constrains design freedom.  But such constraints are not necessarily a bad thing.  For example, analog device designers, though tempted to design custom components, usually have the discipline to accept the boundaries of the rigid process offered by the foundry. This discipline forces these designers to think around the problem and look for an alternative way to achieve the same functionality--and they almost always find it, if only because even considering flexibility in a foundry process would be prohibitively costly.  The same discipline should now be the goal of the MST community, even though the present state of the art is incompatible with the foundry model because of a lack of commonality.

In order to evolve towards such a MST foundry model, the elaboration of a consensus MST taxonomy is a fundamentally important task.

References

1. Boudreaux, J.C.   "Concurrency, device abstraction, and real‑time processing in AMPLE," in W.A. Gruver and J.C. Boud­reaux (eds.), Intelligent Manufacturing: Pro­gramming Environ­ments for CIM, Springer‑­Verlag, 1993; 31-91.
2. Boudreaux, J.C. and Gaitan, M. “Integration Strategies for MEMS Technologies,” Commercializing Exogenous Disruptive Technologies: The Commercialization of Microsystems ‘98; 423-432.
3. Brignell, J. and White, N. Intelligent Sensor Systems, Institute of Physics Publish­ing, Philadel­phia, 1994.
4. Bryzek, J. "Introduction to IEEE-P1451, the emerging hardware-independent commu­nication standard for smart transducers," Sensors and Actuators, A62 (1997); 711-723.
5. Busch-Vishniac, I.J. "Trends in Electro­mech­anical Transduction," Physics Today, July 1998; 28-34.
6. Christensen, C.M. The Innovator’s Dilemma, Harper Collins, 1997.
7. Correia, J.H., Cretu, E., Bartek, M., W­olffenbuttel, R.F. "A local bus for multi-chip-module-based microinstrumentation systems," Sensors and Actuators, A68 (1998); 460-465.
8. Eddy, D.S. and Sparks, D.R. "Applica­tion of MEMS Technology in Automotive Sensors and Actuators," Proceedings of the IEEE, vol 86 (1998); 1747-1755.
9. Fugita, H. "Microactuators and Micro­ma­chines," Proceedings of the IEEE, vol 86 (1998); 1721-1732.
10. Fugita, H., Ataka, M., and Konishi, S. "Group work of distributed microactuators," Robotica, vol 14 (1996); 487-492.
11. Ioannou, P.A. and Sun, J.  Robust Adaptive Control, Prentice-Hall, 1996.
12. Mason, A., Yazdi, N., Chavan, A.V., Najafi, K., and Wise, K.D., "A Generic Multielement Microsystem for Portable Wireless Applications," Proceedings of the IEEE, vol 86 (1998); 1733-1746.     

Biography

Dr. Boudreaux joined the National Institute of Standards and Technology, formerly the National Bureau of Standards, in 1979, first in the Computer Science Laboratory, and then from 1984 to 1992 in the Manufacturing Engineering Laboratory.  In 1992, he was appointed a program manager in the NIST Advanced Technology Program. His research has focused primarily in the following areas: models for higher-order functional systems, adaptive control, real-time error compensation systems, and intelligent manufacturing systems.

Contact information:

J.C. Boudreaux
NIST/Advanced Technology Program
Administration Building, Room A-221
Gaithersburg MD 20899

Telephone (301)975-3560 | Facsimile (301)548-1087
E-mail  jack.boudreaux@nist.gov

Date created: December 10, 2001
Last updated: May 19, 2005