4
INTRODUCTION
Although many factors contribute to the success of any military operation, it has long been recognized that information is one of the most important-information in many different forms and acquired on many different time scales.
Information
During conflict situations several different kinds of information come into play. At the highest top-down level is situation awareness. Warfighters must understand everything that they can about the nature of the opposing forces—their current positions, their movements, their composition, their infrastructure, their capabilities, their communications, their weapons, their threats, their plans—in short, everything. The more the better—in real time on a scale that ranges from minutes to hours—not what they were doing a day ago, but what they are doing right now. Obviously, for maximum cooperative effectiveness the United States ought to have the same complete picture of its own forces, distinguishing accurately between friendly and adversarial forces to minimize or eliminate friendly-fire incidents. To complete the scenario, an accurate, real-time picture of the environment is needed—e.g., the details of the terrain, the current and anticipated weather or sea conditions, the presence or absence of mines, and so on. Generally, in real conflicts, only a few of these factors are actually known to the degree desired—knowledge of some may be stale and out of date, and other factors may only be guessed at and some completely unknown. The better the job U.S. forces do in achieving valid situation awareness, the larger the potential competitive advantage they can enjoy.
On a shorter time scale, from the bottom-up point of view, effective utilization of weapons requires detailed and timely—fractions of seconds to minutes—information about both the targets and the weapons themselves. Targets must be recognized as such, their positions localized instantaneously, their motion measured to high precision, their most vulnerable aim points identified, and so on. Similarly detailed continuous information about the weapons is needed to aim or guide them to a successful interception of target—e.g., weapon position, inertial parameters (such as orientation, velocity, and acceleration), and environmental parameters (wind and tidal currents). On a much longer time scale, outside actual combat situations, information is needed in several forms to provide for such necessities as equipment maintenance and support and overall logistics. For example; it is necessary to know what has failed or is about to fail, or what the weather will be tomorrow.
To acquire desired information, measurements of all kinds of physical parameters must be made. The devices that permit these measurements are known broadly as sensors. The term "sensors" encompasses an enormous range of technologies and devices. Some can be as simple and old fashioned as the direct measurement of a local temperature by means of a thermocouple, and other can be as modern as the detection of a biological agent by a miniaturized mass spectrometer, or as complex as a synthetic aperture radar (SAR) all-weather imaging system. In all cases, whatever the sensor, an interaction between the sensor and its local physical environment results in the generation of some kind of signal, generally a form of electrical response of the sensor's physics, chemistry, and biology to the physics, chemistry, and biology of the outside world. The interpretation of these sensor signals through signal processing, data fusion, and the like leads ultimately to the extraction of the desired information.
Operational Modes
Sensors, whatever information they are attempting to collect, can be broadly classified into two categories of operation, passive and active, which are defined as follows:
Sensors that operate on the basis of wave-like, propagating physical phenomena such as electromagnetic waves, e.g., radar and laser detection and ranging (LIDAR), or acoustic waves, e.g., seismic sensors and sonar, can be operated in either passive or active mode to collect different kinds of information. Sensors based on nonpropagating phenomena such as those that sense chemical compounds or accelerations (initial sensors) operate only passively.
Physical Phenomena
The range of sensor types of interest to naval forces is enormous. Box 4.1
lists the basic physical phenomena that underlie the types of relevant sensors.
As can be appreciated from this list, the subject of sensors is very complex,
involving a large number of apparently quite disparate technical
disciplines.
| BOX 4.1 Physical Phenomena Underlying Sensors
Electric only |
Generic Sensor Model
Each of the physical phenomena listed in Box 4.1 is discussed below in more detail in the context of the current state of the art and projected possible future performance capabilities of the sensor technology associated with each. Although useful for better appreciation of the individual sensors and sensor technologies, this approach can be quite confusing when it comes to projecting the future because of the great variety in sensor types and related details of technology that mask their underlying common features. This section discusses sensing in general and the common concepts and issues that characterize all of sensor technology. Five critical technologies—semiconductors, superconductor, digital, computer, and algorith technologies—appear to be common to all sensor types, and careful delineation of these greatly simplifies the difficulties of projecting future sensor capabilities. Insofar as it is possible, for each identified common critical technology, quantitative performance projections based on today's historical performance growth patterns are discussed below as clues to future potential. A model of a generic sensor is shown in Figure 4.1.
FIGURE 4.1 Generic sensor model.
Sensor Interface with the External World
The first thing that must be considered is the interface between the sensor and the physical phenomena to be sensed. For some classes of sensors, e.g., chemical or biological sensors, physical samples of atoms or molecules or chunks of material must be collected and inserted into the sensor's detection mechanism. The design of this kind of interface permits a good deal of flexibility and no doubt will vary considerably over time.
Sensors such as LIDAR designed for propagating phenomena collect samples that are less material in chracter and more wave-like; e.g., the detected objects are photons or phonons. However, the interfaces for such sensors are highly constrained by the free-space wave-propagating physics of the phenomenon being sensed: Whatever the details of the implementation technology, the sensor interface to the outside world must provide an appropriate wave equation impedance match. The details of the sensor implementations can be altered by the designer, but the outside world's physics is whatever it is and is not under the control of the designer or the sensor. When sensor performance capabilities are projected into the near and far future, the interface constraint remains invariant-e.g., although it will be possible, with time, to compress more and more digital and computing capabilities into ever-decreasing volumes, the dimensions and characteristics of the radar aperture needed for a given task will remain basically the same. Beamwidth requires so many free-space wavelengths across the aperture, and grating lobe suppression requires a certain minimal spacing of elements, again in terms of the free-space wavelength. Certain characteristics of the interface are, and will always be, independent of the technology used to implement the sensor hardware.
For sensors—e.g., inertial, gravitational, and timeas the sensors are totally and inextricably immersed in the phenomena to be sensed, the concepts of external world interface and collection are simply moot.
Detection
Once properly interfaced to the external world, the sensor must selectively detect the manifestations of the phenomena of interest and produce signals that can be used to quantify and convey the desired information. Although in the past many simple sensors used purely mechanical means of indicating the detected signal, as, for example, an automobile thermostat or a thermometer based on the motion of a bimetallic strip or membrane, most sophisticated sensors of interest to the Navy and Marine Corps indicate the results of detection as an electrical signal—a modulated voltage or current. Even though the physical phenomenon being sensed may not be directly electrical in nature, but rather chemical or biological or acoustic, the detection is usually accomplished by using the phenomenon of interest to generate or move free electrons, thereby transducing the physical manifestations into electrical signals. For example, microwave signals are detected by causing the associated time-varying electromagnetic fields to induce currents in electrons already available in conducting elements of the sensor. Optical signals generally use the large energy inherent in each photon (E = hn) to kick loose bound electrons to create free electrons that are then collected and/or moved to produce measurable charges or currents. Other sensors use mechanical components that move under the influence of the physical phenomenon to be sensed but that are also part of an electrical capacitor or other circuit element so that the resulting motion alters the circuit parameters, thereby modulating an electrical signal in an interpretable way. Chemical and biological sensors use yet other techniques to produce electrical detection outputs.
The technology for processing signals that are electrical in form is well understood, and the future capabilities of many diverse sensors can be projected in terms of electrical signal processing technology. Electrical signals are now uniformly dealt with via electronics. Modern electronics is uniformly based on semiconductor technology, and projected progress in semiconductors is often directly translatable into valid projections of improved sensors and sensing. In addition, superconductor technology offers many attractive ways of dealing with electrical signals that operate on the basis of quantum effects that are quite different from those encountered in semiconductors. Although still very much in development, because of the great promise and the recent progress demonstrated, progress in superconductor technology must also be carefully assessed for application to future advanced sensors.
Although electrical signals from independent sensor elements are often combined directly today in analog form, increasingly they are converted immediately to digital format. It is virtually certain that this will be the only approach considered in the future because of the many proven advantages of digital technology. Thus, projected progress in digital technology will be directly translatable into projected improvements in sensors and sensing.
Information Extraction
The stream of digital data emerging from each individual sensor element, e.g., each pixel of an IR focal plane array or each receive element of a phased array radar, must be assembled, stored, processed to extract the desired information, and communicated to a user—sometimes a human and sometimes another mechanical/electronic device— can use this information for guidance and control or some other purpose. Progress in all aspects of computer technology, and particularly algorithm technology, will translate directly into improved, more capable sensors.
That sensor technology, in its myriad
embodiments, is critical to the success of all naval force operational tasks
or missions is obvious (Box 4.2). Whenever information is required, sensors
are utilized to make the physical measurements from which the desired information
is extracted. Radar, optics, and sonar sensors, through the active or passive
exploitation of the physics of wave propagation, give information about distant
objects that is useful for general surveillance and situation awareness as
well as for more specific purposes, such as real-time target location and
weapon guidance. Other sensors, such as position-sensing devices or inertial
sensors, produce useful real-time local measurements that can be used to
control all kinds of platforms, including whole ships, steerable radar or
communication antennas, and gun mounts on ships, or even individual missiles
in flight, depending on just where the sensors are located. Yet other sensors
produce measurements for which the long-term variations in the measured
parameters provide the useful information. For example, temperature or
atmospheric pressure sensors can supply inputs for short- and long-term weather
prediction, whereas acoustic sensors mounted on rotating machinery can provide
evidence of bearing wear or imminent gear failure, thus triggering needed
repair and maintenance procedures.
| BOX 4.2 Sensor-dependent Operational Tasks and Missions Situation awareness General foe/friend information Surveillance Threat detection, recognition, and localization-general or specific Weapons targeting-offensive or defensive Logistics and maintenance |
In short, naval forces are heavily dependent on the use of sensors today, and the future seems to promise even broader use of sensors as the technology continues to evolve toward more capable performance and the demand for more and better information escalates. Future sensors, as compared with existing implementations, promise to cost less, have higher sensitivity and precision, be available in much smaller, lower-power packages, and perhaps be capable of measurements currently unimagined (i.e., be completely new types of sensors).
TECHNOLOGY STATUS AND TRENDS
Despite the dangers in attempting
to project tomorrow's technology entirely in terms of what we see today,
doing so can still impart valuable lessons. Preceding the panel's discussion,
of the five technologies critical to all modern sensors is a brief review
below of the technology trends that are evident in today's developments and
that are shared, in some way, by all classes of sensors (Box 4.3). Toward
the end of this chapter, the panel speculates on the impact of sensor
technologies on tomorrow's naval forces touching on other possible future
directions, not so evident today but desirable from the user's point of view
and not obviously incompatible with some law of nature.
| BOX 4.3 Technology Trends Common to all Classes of Sensors
Solid-state technology |
Solid-state Technology
The most obvious overall trend of significance in technology today is solid-state technology's dominant role in both analog and digital electronics. Today's digital circuits are solid-state—the semiconductor transistor, in one form or another, is the workhorse of the industry and the foundation for all practical digital IC implementations. Its one evident competitor, lurking in the background but always gaining ground, is another solid-state technology—superconductors, e.g., Josephson junctions, superconductor quantum interference devices (SQUIDs), and RSFQ logic.
Even in the analog world, solid-state technology has come to dominate the audio and video amplifiers in entertainment electronics. In the last decade or so, even the generation of microwave power for radar and communication applications has come to be realized increasingly often in solid-state form. Although many legacy microwave systems still generate RF by means of tubes, all new radars are automatically assumed to be some form of solid-state phased array, and soon the only place tube technology will be found is in microwave ovens. Today, magnetrons are still cheaper than the equivalent power transistor, but that may not last as solid-state electronics continue simultaneously to improve in performance and fall in price.
Given this widespread trend, it seems likely that all future advanced sensors will process their detected electrical signals with some form of solid-state circuitry. It can be expected, then, that advanced sensors will share in the continually improving aspects of solid-state technology, that is, increasing miniaturization, higher speeds, decreasing power per function, increasing device density and complexity via IC manufacturing techniques, integral packaging concepts, and decreasing unit costs. A spin-off application of semiconductor manufacturing to three-dimensional micromachining of silicon (Si), enabling the development of MEMS, has already produced a range of novel sensors and actuators with significant performance and cost advantages over the conventional forms.
Atomic-level Manipulation
One of the most exciting recent technology developments is the growing ability to manipulate matter at the atomic level. Largely because of the efforts applied to the fabrication of solid-state devices and integrated circuits, through mastery of thin-film deposition techniques and the physics and chemistry of surface phenomenon, it is now possible to control material and structural fabrication right at the level of the individual atoms. These skills have already been used to create artificial materials with unique characteristics, e.g., alternating-layer semiconductor structures for advanced microwave devices, integrated multilayer Bragg reflectors for photonic applications, magnetic structures with as many as 50 alternating layers of iron and chromium for giant magnetoresistance sensors, biologically inspired self-assembling layered materials of polymers and ceramics with unusual properties, and even artificial high-temperature superconductors with monomolecular layers of alkaline earth atoms (calcium, strontium, or barium) alternating with copper dioxide (CuO2) layers.
These techniques allow the tailoring of materials and devices at the nanostructure level,1 i.e., accurate growth and placement of clusters of a few or a few tens of atoms down to the positioning of single atoms. They will provide completely new sensor materials and the quantum wires, dots, and single-electron transistor devices that are likely to be exploited to continue the long-term growth trends in solid-state electronics into the future for decades to come. It seems clear that sensors and sensor systems of all kinds will benefit from these capabilities, getting continuously smaller and cheaper and more capable with time. The implementation of microscale or perhaps even nanoscale self-contained entities with integrated sensors, computers, and actuators will certainly become possible over the next several decades, and such devices will probably represent a mature, widespread technology by 2035.
Digital Implementations
Another very obvious characteristic of modern electronic technology is the inexorable march toward all-digital implementations. The significant advantages of digital versus analog implementation in terms of flexibility of processing, error containment, and robustness against drift have long been recognized, but cost, speed, and other obstacles have hindered the conversion in many areas. With the performance and costs of digital computers and signal processors improving exponentially with time, i.e., by factors of 5 to 10 every few years, the obstacles are vanishing and the digital implementation of all sensors and sensor systems in the near future appears inevitable.
Analog-to-digital conversion (ADC) technology and its converse (DAC) are currently making such rapid strides in sample rate, number of bits, size, and power, that systems that involve the direct digitization of microwave radar and communication signals at gigahertz rates, with adequate dynamic range and low enough size, power, and costs to be considered practical, are already under development in many places, with field deployments expected in less than 5 years. These kinds of ADC/DAC capabilities, combined with accelerating computational capabilities, will permit the implementation of advanced adaptive processing algorithms, e.g., digital beamforming and space-time adaptive processing (STAP), and effective ATR algorithms, as well as the exploitation of multisensor data fusion techniques. Given the ability to digitize and digitally process almost any radar waveform, it is clear that the time is ripe for the application of digital techniques to other aspects of the system, such as the generation of arbitrarily complex radar transmission waveforms with performance features that go well beyond the simple continuous wave (CW), linear frequency modulation (FM), and occasional phase-coded waveforms that have dominated radar technology to date.
Distributed Systems
As individual sensor and computing elements grow ever smaller in size, power, and cost and simultaneously more powerful, the temptation to combine a large number of them into a super complex of distributed, intercommunicating elements becomes irresistible. The trend today is toward distributed phased-array antennas for radar and communications, multiprocessor supercomputer architectures, distributed power conditioning, the Internet, and so on. And there are enormous advantages to be gained—more information-bearing signals can be collected, more overall computer throughput can be achieved (but always by a factor less than number of elementary processors combined), and much higher overall reliability with graceful degradation characteristics can be obtained. Since it is always possible to make things complicated faster than it is possible to improve the reliability of the individual elements, fault-tolerant redundancy techniques must be explicitly addressed for graceful fail-safe degradation. Equally obvious is the need for efficient, high-bandwidth interelement communications, probably fiber optic and wireless, in which data compression techniques, both lossless and lossy, will come into play.
Moreover, as the Internet has shown, information collected from many, perhaps widely dispersed sources, with the proper communications technology and protocols in place, can be profitably correlated to allow a fuller understanding of a subject or situation. Data-fusion and data-compression algorithms will play a key role in the application of these concepts to sensing and to the achievement of the desired battlefield situational awareness, ATR, damage assessment, and related capabilities. An obvious danger lies in the potential flood of data that a capable multisensor distributed data collection system can create. It is possible to generate overwhelming amounts of data that can cause a total shutdown of the human users, whose performance is notoriously nonlinear and prone to catastrophic collapse. (Degradation is definitely not graceful for overloaded humans.) Concepts and algorithms to permit the recognition, extraction, and viewing of only the minimally required information from large distributed databases must be developed—an area of significant future need for research and development.
Eventually powerful miniature sensors will be combined in a single package with on-board, integral computational capabilities to form mini-systems-on-a-chip. An intriguing prospect for the future is the notion of interacting armies of small, capable, autonomous entities—microrobots that fly or crawl or swim—that combine miniature sensor packages with integrated computers, actuators, power sources, and wireless communication capabilities. These assemblies might be capable of functioning as ant societies do, with each participant acting locally on the basis of mostly local information, and the whole assembly functioning effectively to reach some global goal. This kind of implementation of sensors suggests the possibilities of higher overall performance in surveillance, for example, through adaptive, autonomous spatial repositioning of the individual sensors. The development of single, small, flying sensors of this sort is already under way. Successful artificial societies of this type will require the development of a deep understanding of what the appropriate rules of behavior should be and their implementation in software-another topic for future R&D efforts.
Multidimensional Sources of Information
Yet another striking trend in modern sensor system technology is the use of multiple simultaneous sources of information—e.g., spatially dispersed multiple sensors of the same type or perhaps single-sensor systems operating on multiple spectral bands such as several IR bands, several RF bands, one IR and one RF band, and so on. Dual-band IR focal plane arrays with precise pixel alignment between the band images have been produced, and several RF/IR advanced missile seekers (i.e., multispectral seekers) and optical sensors for satellite platforms that collect pixel-aligned data in hundreds of small (e.g., 10 Å) bandwidths across a large range of the optical spectrum—known as hyperspectral systems—are under development at the present time.
The higher the dimension of the information that can be collected from a pixel or an object, the better the chance of correctly detecting and recognizing it. Clearly, effective data fusion and ATR techniques are needed, and these are already areas of active research. On the other hand, significant increases in computational memory and throughput are required and offer additional challenges on the path to achieving high performance and affordability.
Multifunctional Configurations
The final technology trend of significance to the future growth of advanced sensors and sensor systems is the broad and growing interest in the implementation of multifunctional configurations—that is, sensors capable of performing several different functions via shared hardware. It has long been common to combine the search-and-track function on a single radar by using time-sharing of different waveforms and beam scan patterns, thereby gaining certain implementation advantages in size and perhaps cost over the alternative of building a separate radar for each function. For a number of reasons associated with the available real estate and the growing interest in and need for electromagnetic signature control, there is considerable interest in the Navy in providing multifunction capability in single locations.
It is always easy, in principle at least, to share computer resources between multiple functions. What is often more difficult is to share the interface with the external world and the microwave elements between functions, because many of these functions put quite conflicting requirements on these components and, as was mentioned in the introduction to this chapter, the properties of this interface are strongly constrained by the propagation characteristics of the physics of propagating phenomena, be it electromagnetic or acoustic. For example, radars can utilize time sequencing of the transmit and receive functions to achieve the isolation needed to prevent the high-power transmitted energy from leaking into the sensitive receivers and saturating or destroying them, but microwave communication systems typically require continuous simultaneous transmit and receive and cannot use the time-sequencing approach. Electronic support measures (ESMs) systems, which might also utilize the same or adjacent portions of the microwave spectrum as do the radars and communications, have similar requirements for continuous, sensitive, passive reception and again cannot easily exploit time sequencing.
In addition, if radars of quite different frequencies (say, L- and X- or Ku band) attempt to share the same aperture (such as phased-array elements), extreme linearity and difficult-to-implement high-Q microwave filters in the amplifying electronics and wideband radiating elements are required to prevent deleterious cross-coupling of signals between bands, even if time sequencing can be used to implement transmission/reception isolation. Even more difficult is the problem of combining such widely separated frequencies as conventional RF microwave bands with millimeter waves or either of those with electro-optical systems. The physical implementations suitable for each of these spectral regions differ so much that it is next to impossible to usefully share transmit and receive resources. The amplifiers and filters suitable for one region simply do not work for the others, and often, rather than being able to share resources (particularly at the critical, constrained interface with the outside world), the different implementations conflict. There is always room for ingenuity here, but many of the obstacles are fundamental.
CRITICAL COMMON TECHNOLOGIES
The discussion aboveof the generic sensor model identifies five key technologies as common to all modern sensors and sensor systems and as absolutely critical to their performance potential. Understanding the current state of the art of these individual technical areas and the growth patterns that can be extrapolated will allow reasonably confident prediction of the kinds of performance achievable in the future for the different classes of sensors and the kinds of new naval force applications that might be enabled.
Below, each of the critical technologies (Box 4.4) underlying advanced sensors
in general is discussed briefly, and historical growth curves are presented
where possible. Each technology is extremely important in its own right,
for many applications beyond sensors, and each is discussed also in Chapter
2 of this report.
| BOX 4.4 Critical Common
Technologird |
Semiconductor Technology
Conventional Semiconductors
Semiconductors constitute an enormous topic2 that can be surveyed here only briefly, however critical it is to the future development of sensor technology. Since the invention of the transistor some 40 or more years ago, 3 single-component and at least 10 binary semiconductor materials, as well as a number of tertiary materials (e.g., mercury cadmium telluride [HgCdTe] and aluminum gallium arsenide [AlGaAs]), have been exploited for the implementation of a variety of electronic devices and applications. These single and binary semiconductors, ordered by the magnitude of their bandgap, are listed in Table 4.1. Others not listed here may someday be exploited, including mercury telluride (HgTe), manganese selenide (MnSe), gallium antimonide (GaSb), indium nitride (InN), scandium nitride (ScN), aluminum nitride (AlN), zinc selenide (ZnSe), and boron nitride (BN). Also indicated in Table 4.1 is the nature of the bandgap, that is, whether it is direct (D) or indirect (I) gap material, as this determines to a large extent just what kinds of applications the semiconductor may be suited for.
| Material | Bandgap (eV) | Bandgap Type |
| InSb | 0.230 | D |
| InAs | 0.354 | D |
| Ge | 0.664 | I |
| Si | 1.124 | I |
| InP | 1.344 | D |
| GaAs | 1.424 | D |
| CdTe | 1.475 | D |
| AlAs | 2.153 | I |
| GaP | 2.272 | I |
| ZnTe | 2.394 | D |
| SiC | 2.416 | I |
| GaN | 3.503 | D |
| C | 5.5 | I |
First to be exploited, germanium was soon replaced by silicon, clearly the most widely used semiconductor material so far and likely to remain so for the foreseeable future. The reasons for silicon's dominant position are numerous: It is readily available in large quantity; it is mechanically strong; it is a good thermal conductor; it can be easily grown into large-diameter, ultrapure, defect-free crystals; it forms stable insulating oxides of excellent quality; and it is nontoxic and easily fabricated, via a wide range of patterning, etching, implanting, and diffusing techniques, into devices with literally millions of circuits per chip and hundreds to thousands of chips per wafer with high yields.
These virtues, and an enormous multidecade investment in time and resources, have led to the explosive proliferation of digital and microelectronics fabrication technologies that characterize and enable the rapid growth in computer and information technology. Linewidths continue to decrease exponentially with time, with optical lithography still performing effectively at submicron dimensions that were thought to be beyond its capability only a few years ago, and with finer, although less convenient, x-ray and electron-beam techniques waiting in the wings to continue the fabrication down into the regime of quantum dots and wires and ultimately to single electron logic structures—about as far as can be imagined today. With finer dimensional capabilities in hand, and as defect densities continue to be reduced, the number of circuits that can be placed on a single chip with reasonable yield grows exponentially with time, causing the cost per operation to spiral downward while performance, in terms of clock speeds and throughput, continues its exponential upward growth—a pattern of factor-of-10 improvements every 4 or 5 years, which has been consistent for at least a decade and a half and shows no signs of slowing as yet. Figure 4.2 illustrates the exponential growth in the total number of transistors on a single chip from 1970 to the present and also extends the average observed growth pattern to the end of this study's time frame, 2035. Since there are no obvious fundamental physical laws that limit the number of transistors achievable per chip, and given the almost 30 years of observed consistent exponential growth, extrapolation of the observed pattern into the future appears to be reasonable. There is little doubt that this prediction will be quite accurate for the near future, say, the next 5 years, but it is obviously far less certain for the distant future. The obstacles to further growth are generally practical rather than fundamental; e.g., because of material absorption, ultraviolet imaging systems are difficult to implement via conventional optical concepts, until the predicted fabrication linewidth dimensions reach interatomic dimensions near 2035, as is discussed below. With sufficient motivation these obstacles may be overcome and the extrapolated pattern may continue into the future much further than may be evident today.
FIGURE 4.2 Historical trend in silicon technology (following Moore's Law). SOURCE: Data points from Yu, Albert, 1996, "The Future of Microprocessors," IEEE Micro, 16(6):46-53, Figure 3, December.
For all its virtues, silicon is not a perfect semiconductor material for all applications. Its bandgap is rather small, limiting its performance at elevated temperatures. Its bandgap is indirect, which inhibits its use as a laser or light-emitting diode (LED) source and makes its ability to absorb, and hence detect, optical photons weak near the band edge energy; this does not mean that silicon cannot be used as an optical detector but rather that it requires a larger thickness than direct bandgap materials for the same detection effectiveness. Finally, the charge carrier mobility and saturation velocities are rather low for silicon compared with some of the other semiconductors, such as gallium arsenide (GaAs), indium phosphide (InP), silicon carbide (SiC), and carbon (C-diamond), seriously limiting the speeds at which silicon devices can operate.
A number of approaches have evolved to address these limitations, some of which are mature or close to maturity and others less so but which show great promise for the future. Other semiconductor systems offer different inherent physical parameter characteristics that can be exploited, as is discussed below, and alternate device configurations, difficult to implement in silicon, are possible in other technologies with a variety of specific advantages.
After silicon, GaAs is the next most mature semiconductor device technology. GaAs transistors enjoy distinct advantages with respect to speed and power over silicon transistors, because of the shape of GaAs's electron velocity versus electric-field curve-an inherent property of any semiconductor material determined by the specifics of its crystal structure and the resulting band structure. The peak electron velocity in GaAs is several times greater than that in silicon and is reached at a much lower value of electric field.
GaAs field effect transistors (FETs) are capable of operation at clock rates as high as 5+ GHz in digital circuits and as analog amplifiers with significant power and gain in the microwave range up to 30+ GHz. High-performance analog-to-digital converters capable of 6 to 8 bits at sample rates of 2 to 3 × 109 per second have recently been implemented in GaAs heterojunction bipolar transistor technology. And for the past decade or so, solid-state microwave developments have been dominated by monolithic microwave integrated circuit (MMIC) technology—based almost exclusively on GaAs.
In addition to its electron mobility-related advantages, GaAs offers a direct bandgap, making it suitable for optical applications as a detector or as a light source, such as an LED or a laser, thereby easing the interface between its microwave and high-speed digital capabilities and the fiber-optical communication links that will be utilized for the transfer of data from some sensors to their associated computational resources. The direct bandgap also helps in that it produces short minority-carrier lifetimes such that undesired electron-hole pairs recombine so rapidly that parasitic effects may be safely ignored in the design of devices leading to simpler structures.
Of course, GaAs, for all its performance advantages, is a much more difficult material system to work with than silicon. Its hole mobility is low, and there is no natural insulating oxide, so that it is not practical to implement the CMOS structures that offer such low-power performance advantages in silicon. Its dielectric constant is almost twice as large as that of silicon, giving it a higher capacitance for the same-area device. Finally, its low thermal conductivity requires very thin substrate thickness for device thermal control, which, when combined with its extreme brittleness, leads to significant fabrication yield losses through handling. Further yield problems are inherent in the fact that as a more complex system than silicon, it is much more difficult to prepare large-diameter, defect-free GaAs substrates.
Bandgap and Heterojunction Engineering
Two of the most significant concepts to emerge from the efforts expended over the past several decades on the development and maturation of GaAs semiconductor device technology are bandgap engineering and heterojunction engineering, both reflections of our increasing abilities to manipulate materials and structures at the fundamental atomic level. The GaAs high-electron mobility transistor incorporates both of these concepts. The bandgap of complex multicomponent materials, such as AlGaAs, varies continuously as the proportions of components are changed. In this way it is possible to tune or engineer the material's bandgap to optimize transistor performance or, if the device is an LED or a laser diode, to control the specific wavelength emitted. Through the use of molecular beam epitaxy, metal-organic chemical vapor-phase deposition, and combinations thereof, very thin layers of high-quality, defect-free, properly tuned AlGaAs can be fabricated with electron mobility as much as three orders of magnitude larger than can be achieved in silicon and, when grown on ordinary GaAs, produce beneficial heterojunctions at the interface because of the different bandgaps of the two materials. The resulting heterojunction devices show significantly higher switching speeds as digital circuits and improved power and efficiency performance at high microwave and millimeter-wave frequencies than the more straightforward metal-semiconductor field-effect transistor (MESFET) implementations.
These concepts apply broadly to other material systems, and, in fact, their application to silicon through the bandgap and heterojunction engineering of silicon germanide (SiGe3)has produced devices that perform as well as or better than GaAs devices but retain much of the simplicity and ease of manufacturing aspects of pure silicon technology. Indicative of the continuously accelerating growth of technology today, SiGe has moved from a laboratory curiosity less than a decade ago to a commercial reality today. SiGe transistors have already demonstrated cutoff frequencies in excess of 100 GHz, which is better than the best silicon transistors by more than a factor of two and in the same ball park as GaAs and InP technologies. SiGe is an exciting and promising technology and could someday displace GaAs completely.
Next of interest in the III-V family for electronics and MMIC applications is InP. Although behind GaAs in development, InP exhibits even higher electron mobility characteristics than GaAs and also has a direct bandgap that is somewhat smaller than that of GaAs, but just right to permit bandgap engineering of LEDs and laser diodes in the two wavelength bands of most interest to long-distance fiber-optic communication applications (i.e., 1.3 mm for minimum dispersion and 1.55 mm for minimum loss). High-electron-mobility transistor(HEMT) or pseudomorphic HEMT (PHEMT) devices for millimeter-wave applications have shown higher power and efficiency than similar GaAs implementations and are very promising. Unfortunately, InP has many of the same negative features as GaAs technology and currently lags it in overall maturity.
Wide-bandgap Semiconductors
Even more interesting, and still further behind in maturity, are the so-called wide-bandgap semiconductors (WBSs). Although this term is often not used precisely, it generally includes the IV and IV-IV materials and the III-V nitride compounds.4 The most promising of these are SiC, gallium nitride (GaN), and carbon (diamond)5, all with bandgaps above 2.4 eV. Because these materials have such large bandgaps and much higher thermal conductivity and dielectric breakdown strengths than silicon, GaAs, or InP, they offer promising high-temperature and high-power performance. With low-dielectric constants and high-electron mobility, the prospects for millimeter-wave capabilities are excellent. And as these materials are much harder and stronger than conventional semiconductors, they offer the potential for higher processing yields and lower manufacturing costs. In addition, SiC, GaN, and diamond all offer the potential for blue, green, and ultraviolet light emission and have been proposed for application as LEDs and radiation detectors (visible and ultraviolet) as well as for power and microwave devices (bipolar, MESFET, and IMPATT), thermistor sensors, and high-speed switching devices.
Although SiC was one of the first semiconductors recognized (electroluminescence was reported in 1907), and the electronic properties of diamond were first investigated in the 1930s, most of the progress in WBS technology has come in the last decade, and formidable fabrication obstacles remain. Detailed predictions of SiC and C (diamond) microwave devices have been made by means of basic measured material parameters and device models that correctly predict observed GaAs device performance. Figure 4.3 illustrates predicted performance for SiC and diamond MESFETs compared with measured performances of equivalent GaAs and silicon devices. The GaAs results represent the state of the art for GaAs MESFETs. The predicted performance of SiC and diamond MESFETs is significantly better than that for GaAs, suggesting that at 100 GHz, about 300 mW and 1 W of RF power is possible from SiC and diamond devices, respectively. At lower frequencies, in the more traditional microwave bands, significant power performance is anticipated, i.e., from CW power amplifiers with tens of kilowatts at L-band and below, to 1 kW at S-band, several hundred watts at X-band, and several tens of watts at 35 GHz (Ka band).
FIGURE 4.3 RF power performance vs. frequency for diamond, SiC, and GaAs MESFETs. SOURCE: Adapted from Trew, Robert J., Jing-Bang Yan, and Philip M. Mock, 1995, "The Potential of Diamond and SiC Electronic Devices for Microwave and Millimeter Wave Power Applications," Proceedings of the IEEE, 79(5):598-620, Figure 15. Copyright 1995 by IEEE.
The obstacles that remain are related to achieving high-quality, uniform films with controllable properties. The technology for producing single-crystal films of diamond with understanding and control of nucleation, growth, the methods of impurity introduction and activation, and the formation of ohmic contacts with good adhesion is still at an embryonic stage and must be mastered before diamond electronics can become a reality. SiC and GaN materials and fabrication technologies are in better shape and, although GaN is still in a basic research phase, a host of practical devices have already been implemented in SiC. There is little question that, via WBS, high-power, high-temperature, robust semiconductor electronics with interesting potential for short-wavelength optical applications will become available in the next several decades. The growth of microwave power generation over the past decade and a half is illustrated in Figure 4.4, showing the increase in X-band amplifier power performance.
FIGURE 4.4 X-band power amplifier output. SOURCE: Adapted from the U.S. Air Force Scientific Advisory Board, 1995, New World Vistas, Air and Space Power for the 21st Century, Sensors Volume, U.S. Air Force, Washington, D.C., Figure 6-6, p. 102 and data are shown in Figure 4.4 of this report.
Extrapolation of these curves into the future and adding the levels of power capability predicted for SiC and diamond suggest that SiC may represent the state of the art by about 2005 and that diamond needs to be fully mature by 2010 to 2015 to be the state of the art at that time. Further progress may simply involve multiple MESFET amplifier chains on a single substrate. With SiC and diamond, it should be possible to maintain the exponential trend out to 2020 and beyond. Of course, the real art at that time may come from technologies not envisioned today—the details are rarely predictable, but it is highly probable that the envelope will persist.
Higher Levels of Integration
Digital Circuits
Semiconductor transistors, particularly those designed for high frequencies, are by nature quite small—the active dimensions of single transistors, whatever their design, are limited by fundamental physical properties of the materials, e.g., electron and hole mobility, to dimensions typically measured in microns to tens of microns. Combining this feature with the fact that the fabrication techniques employed can frequently be applied uniformly over wafers of many inches in diameter leads to the possibility of fabricating large numbers of transistors on the same small substrate at the same time. In the digital world, this rather obvious extension from single devices to integrated circuits (ICs) took some time to evolve. The first key was to understand how to interconnect the active devices with useful passive components, equally small, that could be fabricated by the same photolithographic, deposition, diffusion, and etching techniques that produced the transistors. The second was to control the fabrication imperfections so that economical production yields of fully functional integrated circuits could be obtained. The final challenge lay in the ability to design and simulate accurately, through software and CAD tools, complex circuits with hundreds to thousands to millions of devices. This improved design and simulation capability was very much a bootstrap operation, as each generation of computer chip enabled the more powerful software tools needed to implement the next generation.
For silicon digital ICs, these endeavors have been quite successful, as is well illustrated by the silicon-technology growth curve (see Figure 4.2) presented earlier. For other semiconductor systems, particularly those with multiple components, although the fabrications techniques are similar to those used for silicon, the systems are generally more difficult to control, and so the maturity of these technologies is significantly behind that of digital silicon. For example, the closest contender, digital GaAs, which has many performance advantages over silicon in speed, in particular, is characterized today by only tens of thousands of gates per IC capability, whereas state-of-the-art silicon technology can produce several million transistors per chip circuit with economical yields. No doubt, some time in the future, when silicon technology's capabilities are saturated-even with quantum devices and single-electron transistors this will happen-other semiconductor systems, such as GaAs, SiGe, and SiC, will catch up and continue the overall digital technology growth envelope illustrated above.
Analog Circuits
Although the same arguments for a high level of integration implementation apply to microwave and optical analog circuits, the obstacles have proven to be different in detail and much more difficult. Although microwave transistors, optical detectors and emitters, and various passive components, including solid-state strip line and waveguide transmission line structures with low loss and good impedance control, can be made by the same microelectronics manufacturing techniques as used for digital electronics, these high-frequency applications cannot approach the level of integration that characterizes digital devices. Not only do these high-frequency applications demand more precise control over dimensions, impedance, and losses, but also the passive components required are physically much larger than those used by the digital implementations. For although integrated, both microwaves and light must propagate finite distances before useful operations can be performed on them by passive structures, e.g., RF inductors and Mach-Zender modulators. Combined with the more difficult multicomponent semiconductor systems, such as GaAs, InP, AlGaAs/GaAs, and InGaAsP/InP, which characterize these applications, the result is that microwave and photonic chips are inevitably limited to device-per-chip densities that are a small fraction of what can be implemented in modern digital silicon technology, i.e., tens to hundreds of components per IC rather than the thousands to millions that characterize today's digital chips.
In spite of this unpleasant obstacle, the microwave and optical ICs are as capable and far smaller and lighter than their conventional equivalents that use discrete components and free-space propagation and, because of the monolithic processing used, offer many cost and reliability advantages. This combination of advantages has already opened up new application areas, the enabling of phased arrays by MMIC technology being a good example. Without doubt, the trend toward smaller size and more capability per analog chip will continue into the future.
Microwave Components
During the past several decades, great progress has been made in applying HF semiconductor technology to the generation of MMIC for a wide variety of useful radar and communication applications—active antenna phased arrays, in particular. MMICs are integrated circuits containing multiple active devices as well as integral passive components such as diodes, resistors, capacitors, inductors, and low-loss controlled-impedance transmission lines, and they perform useful microwave functions such as low-noise amplification (LNA), power amplification (PA), phase shifting, and attenuation. It would be desirable, from a manufacturing cost and perhaps a performance point of view, to be able to fabricate multiple functions on a single MMIC chip—such as a complete transmit/receive module with oscillators, filters, mixers, PA, LNA, circulators, and the like. For many reasons, this is not practical today, and so current practice combines limited-function MMIC chips in a hybrid package, very much like a digital multichip module (MCM), which implies costly, difficult to automate, and often unreliable discrete interconnections from chip to substrate to chip.
The application of MMIC technology to phased-array radar and communications is limited by the cost of individual transmitter, receiver, or transmitter/receiver (T/R) modules. Current efforts are largely devoted to reducing phased-array element costs, i.e., the costs of the T/R MMIC plus the support structure, cooling, radiating elements, and such. To make phased-array radar and communications applications affordable, these costs must be reduced by one or two orders of magnitude. Currently, the total antenna cost divided by the number of phased-array elements ranges in the thousands of dollars. The best approach to affordability would seem to lie in mastering the implementation of single-chip MMIC modules so that low-cost, automated microelectronic manufacturing practices can be applied. This should occur as the technology continues to evolve.
For microwave applications, achieving the highest level of complexity and power-generating capability in the smallest volume possible is not always desirable or necessary. Once a phased-array element can be made small enough to fit the element separation constraints and still give the required performance, it makes little sense to try to reduce its size further if there are no accompanying significant benefits in power, weight, or cost. On the other hand, it may make sense to add additional functional capabilities to enable multifunctional performance (i.e., different functions from the same physical aperture), but this suggests only modest increases in complexity, that is, by the number of separate functionalities needed.
From another point of view, as is discussed below, digital techniques will inevitably move as close to the external interface (antenna) as possible, suggesting the possibility of combined microwave-digital chips of great complexity but with the complexity largely confined to the digital portions. Such hybrids have already been implemented, i.e., MMIC chips with integral GaAs digital control logic on board, although the fabrication of analog and digital transistors differs in details, e.g., impurity profiles, that are often incompatible. Appropriate compromises and techniques will be found, and progress will continue in this direction.
Optical Components
Detectors—Focal Plane Arrays
Semiconductors form natural optical detectors, because incident photons whose energy exceeds the bandgap (huEBG) readily kick electrons from the valence band up into the conduction band, giving rise to measurable electrical responses. The wavelength corresponding to the bandgap is known as the cutoff wavelength (lcutoff), as all radiation with wavelengths less than the bandgap wavelength will have enough energy to generate a response and wavelengths that are longer produce no response at all. To reduce the effects of thermal noise inevitably present in any semiconductor electronic circuit, the temperature of the detector (TD) must be kept low enough so that KTD is well below the photon energy hucutoff. In practice, this reduces the necessity to satisfy the approximate relationship, TD lcutoff~ 550 K mm. Thus detectors for the 3- to 5-mm IR spectral region must be cooled to about 110 K, which, in practice, implies liquid nitrogen at 77 K. For longer IR wavelengths, say out to 20 mm, 30 K is required, which demands cryo-engine cooler technology. For semiconductor optical detectors, thermal control can be quite a limiting and expensive inconvenience. Fortunately, for the near-IR (<1.8 m), visible, and UV regions, optimal room temperature operation is feasible.
Unfortunately, even with appropriate cooling, there is no single, ideal, detector material that gives optimal performance for all IR, optical, and UV spectral bands of interest to sensing. And so, over the years, many different semiconductor detector systems—e.g., various forms of silicon and germanium with different doping for different spectral regions, lead sulfide (PbS), lead selenide (PbSe), indium antimonide (InSb), platinum silicide (PtSi), various compositions of HgCdTe for different spectral regions, and so onhave been employed. Each system requires its own fabrication techniques and today exhibits varying levels of maturity, depending on the specific system. It is no surprise to find that the visible and near-IR regions are currently the most mature as they are typically implemented in silicon.
Although single optical detectors find use in some nonimaging situations, the application of most interest is imaging. The earliest imaging systems employed single detectors in scanning configurations, but with the development of silicon microelectronic IC technology, visible focal plane arrays (FPAs) with multiple silicon detectors in one- and two-dimensional configurations were rapidly developed during the early 1980s. By means of traditional optics, an image is projected on the focal plane where the detector elements respond in parallel, accumulating charges proportional to the image irradiance levels at each detector position. The signals are then read out very rapidly, serially or simultaneously, into some form of readout electronics, for display or storage. The detector elements are then reset and the cycle repeated, typically at video frame rates of 30 Hz or higher. Conveniently, the readout circuitry, often a CCD configuration, could also be implemented in silicon and integrated directly on the FPA itself to form an all-solid-state monolithic imaging detector. These, of course, form the basis for the high-quality and inexpensive video camcorders so widely available today. Sensors that operate in the visible region with millions of picture elements (pixels), e.g., 2,000 × 2,000, are in production now, and formats with several tens of millions of pixels, e.g., 5,000 × 5,000, will soon be possible.
Applying these same FPA concepts to the semiconductor materials needed for optimal performance in the longer IR bands, such as InSb, PtSi, or HgCdTe, is possible but is far less straightforward. Generally, in these technologies, implementing the readout circuitry on the same substrate with the detector elements in a monolithic form is not practical, and so the detector array must be interfaced to an external silicon very large scale integrated (VLSI) circuit chip. In addition, as is the case for digital circuits, it is much more difficult to economically produce large-area structures in these more exotic systems. And so, medium-wavelength IR (MWIR), i.e., 3 to 5 mm, and long-wavelength IR (LWIR), i.e., 8 to 14 mm, FPAs lag the visible somewhat, with MWIR FPAs of 1,024 × 1,024 elements in InSb and HgCdTe and LWIR FPAs of 256 × 256 elements in HgCdTe and extrinsic silicon representative of the current state of the art.
Optoelectronics
Semiconductor optical capabilities are by no means limited to the detection of light but also can be readily configured to generate light at the bandgap energy—that is, at lcutoff, as LEDs or as laser diodes. With the advanced fabrication techniques, such as molecular beam epitaxy (MBE), metallo-organic chemical vapor deposition (MOCVD), ion beam implantation, and micromachining, miniature high-quality optical waveguides, mirrors, and lenses can be manufactured. Several decades ago these possibilities gave rise to the concept of integrated optics, that is, optical systems that could be implemented in a miniature, monolithic, solid-state form. Although significantly smaller than the conventional free-space optical alternatives, integrated optics implementations are never as small as digital ICs, as the name suggests, because of the necessity of using finite amounts of optical propagation distance to achieve useful amounts of modulation, mixing, diffraction, and so on. The dimensions of integrated optical elements are commonly measured in millimeters or centimeters, rather than in microns, as is the case for digital ICs.
As very thin layer, multiple quantum well (MQW) two-dimensional electron gas and tunable bandgap engineering concepts evolved for the optimization of single transistor performance, these concepts were quickly applied to the generation of efficient LED and laser diode structures for many different spectral regions. The simple discrete laser diodes of the 1960s, with optical feedback provided by polished or cleaved reflective facets at the ends of the active p-n junction region, gave way to sophisticated MQW multilayered structures with gradient index and multilayer distributed feedback (DFB) confinement built in. To meet the needs of long-distance fiber-optics digital communications—the most successful application of optics to electronics so far—CW room-temperature semiconductor lasers, i.e., InGaAsP/InP MQW devices, capable of direct current modulation with bandwidths of tens of gigahertz and with outputs approaching 1 W or so, have been developed for both the 1.3-mm (minimum dispersion) and the 1.55-mm (minimum attenuation) spectral regions of silica fibers.6 Adding strained layer concepts promises to further improve the power, efficiency, and lasing thresholds of MQW lasers in the future. Because of the growing fiber-optics communication business, commercial sources for 1.3-mm and the 1.55-mm lasers abound.
In addition to long-distance fiber-optic digital-communications applications, there is increasing interest in the possibility of using optics to transfer analog microwave signals within phased-array radars to supply time-delay steering, as well as to distribute the high-data-rate, large volumes of digital data associated with modern digital receiver radar concepts. For these applications, the earlier-developed and simpler AlGaAs/GaAs MQW lasers, which operate in the 0.8-mm spectral region, are ideal, as the increased fiber attenuation and dispersion in this spectral band are of no consequence for the minuscule distances (a few to a few tens of meters) typical of a radar, and the commonality with the GaAs MMIC technology suggests the possibility of fully integrated, monolithic optical-microwave implementations. Interest in these applications should grow in the near future, although the technology of AlGaAs/GaAs MQW lasers has languished up to now in view of its inapplicability to long-haul communications, and there are no commercial sources for these lasers currently available.
The possibilities of combining optical and electronic functions on the same chip bring us finally to the broader subjects of optoelectronics7 and the optoelectronic integrated circuit (OEIC). The formidable manufacturing technology that has evolved to support digital and microwave microelectronics has enabled optical, electro-optic, and electronic components to be implemented together in monolithic semiconductor form, combining the advantages of each normally disparate element in a single device—the OEIC. Implementing drive electronics on the same chip with a laser diode makes an attractive cheaper, faster, smaller fiber-optic transmitter for communication applications and was one of the first OEICs developed. Equally important are the fiber-optic receivers, and it is natural to combine detectors with transimpedance low-noise amplifiers as an OEIC. For long-distance communications, this implies InP digital/analog electronics, whereas for photonic radar applications, GaAs technology is suitable. OEIC chips in GaAs, combining high-bandwidth optical detectors with both matched digital electronics and MMIC amplifiers suitable for extracting digital and microwave information transmitted simultaneously on a single optical carrier, have already been demonstrated in the laboratory. Interest in this technology for the implementation of photonic radars with exceptional properties is running high at the present moment, but formidable practical obstacles still remain, and no photonic-based radar has yet been fielded.
Although the success of the application of optoelectronics to microwave systems is uncertain, there is an area of application that is just building momentum but that seems destined to succeed in the end—free-space photonics for broadband interconnects, e.g., chip to chip or board to board, internal to computers. Progress in computer technology will be of great interest to future sensor systems. Analysis suggests that, for distances of more than about a millimeter and data rates of more than 100 Mbps, less power is required on the chip for photonic interconnects than for electronics. Recent progress with very-low-threshold lasers suggests that the break-even distance may soon be only fractions of a millimeter.
Finally, it has been suggested that optoelectronics,8 in the form of various device technologies, such as AT&T's self-electro-optic effect device (SEED), offers the potential for logic implementations that operate at the speed of light and that may someday replace the existing VLSI circuit technology on which the digital information revolution is now based. Although the ability of such optical devices to perform logic has been demonstrated, only very simple functions were implemented, i.e., a few gates. Intriguing as this seems, there are good reasons to believe that conventional digital logic cannot be matched by optical logic, the speed of light notwithstanding. The primary problem is that it takes a good deal of effort to generate, modulate, and detect light—which is always done by moving electrons first, as light does not interact strongly with anything else-and this investment must therefore return significant benefits to be worthwhile. For the communication of wideband information over macroscopic distances, whether these be millimeters or miles, the investment is advantageous and optical communications is the preferred path of the future. For implementing logic, however, particularly in view of the shrinking dimensions of digital circuit elements and with the possibility of single-electron transistors on the horizon, optical logic seems to require the use of many more electrons for each logic operation than is, or will be, required for the digital alternative.
Microelectromechanical Systems
The final semiconductor topic of interest to the future of sensor technology is MEMS. MEMS technology is an imaginative, but logical, exploitation of microelectronics. Through the use of traditional silicon fabrication techniques, microelectronic circuits and miniature, movable mechanical components with dimensions measured in microns are combined on a single substrate to perform a wide range of sensing and actuation tasks (Figure 4.5). Drawing on the fabrication techniques and materials of microelectronics as a basis, MEMS processes are used to construct both mechanical and electrical components. Mechanical components in MEMS, such as transistors in microelectronic devices, can be fabricated with features that have micron or submicron dimensions, thus enabling the inclusion of millions of mechanical components on a single chip. From the mechanical point of view, silicon is ideal for this application because it is strong and easily fabricated into ridges, valleys, free-standing bridges and cantilevers, cavities, nozzles, membranes, and other structures.
FIGURE 4.5 MEMS technology roadmap. Adapted from a briefing by K.J. Gabriel, DARPA, to the NSB Panel on Technology. SOURCE: Adapted from Gabriel, Kaigham J., 1996, "MEMS Technology Trend and Roadmap," in the briefing Microelectromechanical Systems (MEMS), to the Panel on Technology presented by DARPA, September 13.
MEMS technology combines the advantages of miniaturization, multiple components, and large-scale integration. Because of the flexibility inherent in the integration of electrical and mechanical components, the potential applications of MEMS technology appear to be limitless. Examples of possible applications include miniature inertial sensing and guidance devices, miniature sensors of all types that can be widely dispersed or gathered into large arrays, and miniature actuators, including steerable mirrors for directed-energy weapons and smart skin structures for aerodynamic flow control. Other specific military uses include inertial guidance for munitions, integrated fluidic systems for biological and chemical analytical instruments and for hydraulic and pneumatic control, miniature DNA detection systems, integrated micro-optomechanical components for displays, IR detector arrays, fiber-optic switches, and vibration sensors for condition-based maintenance. Some commercial applications of MEMS technology are already available:
Although MEMS technology is being actively developed for commercial applications, the defense community cannot rely on the commercial sector to address all of its development needs because MEMS devices are highly application specific.
Superconductor Technology
High-temperature Superconductors
Superconductor technology has shown tremendous potential for application to both ultralow-loss, high-Q microwave devices9 and to very-high-speed, very-low-power digital circuits—advances that could be incorporated into advanced sensors in the near future through the maturation of high-temperature superconductor (HTS) technology. HTS systems, such as YBaCuO,10 have superconductor critical temperatures well above that of liquid nitrogen (77 K) and do not require the liquid helium (at 4 K) of the earlier classical systems. Because the superconductor phenomenon is a macroscopic manifestation of quantum mechanical behavior, its properties are often strikingly different from what classical physical intuition suggests. These significant differences in behavior permit the implementation of completely new devices—such as SQUIDs, which have found application in accurately sensing weak magnetic fields—as well as the offering of alternate implementations of familiar devices, e.g., microwave components and digital logic circuits,11,12 with completely different characteristics, often permitting very-low-power, very-high-speed performance that cannot be obtained in any other way. However, the obstacles to the realization of this potential have proven formidable and with a few exceptions, medical imaging being a noteworthy example, HTS systems have yet to find widespread, practical, commercial application. Contrary to the popular image, the low temperatures required, although sometimes stressing and awkward with the earlier systems, are much less problematic with HTS technology and the newer generations of cryocoolers than are the basic materials and large-scale fabrication issues that remain.
High-performance Microwave Devices
The best known characteristic of superconductivity—the direct current (dc) resistivity of the superconductor abruptly vanishing as the temperature is lowered below the critical temperature—suggests that very-high-Q, i.e., low-loss, superconductor implementations of microwave delay lines, resonators, and filters may be feasible—and this is indeed true, although it not nearly as straightforward as might be thought. Microwave losses do not vanish below the critical temperature but do diminish rapidly with temperature and vary with frequency, with the losses diminishing toward the dc limit as the frequency is lowered. The combination of low surface resistance and a frequency-independent penetration depth allows the implementation of compact microwave designs with low insertion loss, large bandwidth, and low dispersion. In the past several years, practical thin-film HTS implementations of dielectric resonators with Qs exceeding 3 million and narrow, tunable high-Q microwave filters, operating at 77 K and capable of handling tens to hundreds of watts of power, have become available. Small-scale integration of receiver front ends and other microwave assemblies, with a few tens of superconductor and semiconductor components on a single wafer, also have been demonstrated but are not yet widely available. Such components, particularly the high-Q tunable filters, are certain to play a role in systems that utilize digital receiver techniques with conversion at microwave frequencies.
High-performance Digital Circuits
The full realization of the quantum nature of superconductivity in the late 1950s and the discovery of the Josephson effect in the early 1960s led quickly to the exploitation of the quantized nature of the Josephson junction (JJ) response in the form of digital logic. What JJ technology offered for digital implementations was primarily intrinsic switching speed of only a few picoseconds and low signal voltages determined by the superconductor bandgap, i.e., about 3 mV for traditional low-temperature superconductors, which imply power dissipation per junction of less than 1 microwatt.
However, early attempts to develop superconductor computers at IBM were frustrated by material problems associated with temperature cycling failures and by the relatively low speed, i.e., a few gigahertz, latching concepts utilized. In 1985, Russian workers developed an alternative form of JJ logic, known as rapid single-flux quantum (RSFQ) logic,13 which is much faster, promising 100- to 300-GHz performance capabilities. Rather than switching between a superconducting and a nonsuperconducting state, which the earlier logic used, RSFQ switches superconducting currents between alternate paths, never passing out of the superconducting regime, thereby avoiding the finite transient times associated with the superconducting-to-nonsuperconducting transition. The general concept is familiar from the earlier days of emitter control logic (ECL) semiconductor logic, which utilized the same trick to achieve high-speed performance in exchange for the higher powers associated with never entering the current-off state.
Today, digital circuits and analog-to-digital converters of the RSFQ logic family are under investigation in several laboratories. RSFQ demonstration projects have been built with classical low-temperature materials (superconducting) and, with progress in HTS technology, certainly could be developed into practical special-purpose processors in the near future. A wide variety of ADC of both the Flash and Sigma-Delta14 architectures have been proposed and implemented to some extent. The first superconductor Sigma-Delta ADC was demonstrated last year, in a niobium-based (low-temperature) technology and obtained a respectable 78-dB spurious-free dynamic range bit over a signal bandwidth of at least 5 MHz, through a single-loop modulator clocking at 45 GHz. Performance far better than this is projected for the future.
At present, the most difficult barriers to widespread digital application are the need to fully master the material properties of the HTS systems, which are complex, multicomponent ceramic materials, and the need to adapt the fabrication processes to larger scales of integration than have currently proven practical—a few hundred to a few thousand on a single chip—even in the better-behaved low-temperature niobium systems. Most device demonstrations to date have utilized the low-temperature systems, which remain awkward because of their requirement for more complex cryogenics and are thus less attractive than HTS.
Digital Device Technology
The growth of digital device technology is the single most important factor controlling the foreseeable future of sensor technology, whether it be to the year 2000 or to 2035. The advantages of a digital, rather than an analog, representation for signals or information are many and well known, including immunity from drift, containment of errors introduced by analog-to-digital (A/D) quantization and subsequent digital processing, simpler fabrication requirements (i.e., transistor linearity is not a serious issue), low-power physical implementations, and complete flexibility in defining and changing the processing algorithms, whether they are linear or nonlinear. The march toward a completely digital world has been inexorable over the past several decades. Although most evident today in such devices as the popular cellular telephones, high-definition television (HDTV), and personal computers, the digital revolution is by no means limited to these obvious applications, as is illustrated by the fact that our watches have long been digital, most medical thermometers are now digital, and automobiles, which all run far better and far more reliably than they did 20 years ago, contain multiple embedded digital computers. There is no doubt that all sensors will be digital in the near future—immediately converting the outside world's physics-induced electrical sensor responses to bits for further processing, storage, communication, and display.
Microelectronics
The fundamental characteristics of the growth of digital electronics are most robustly described in terms of integrated circuit fabrication and performance parameters, independent of the semiconductor material systems, the device designs, or the circuit architectures employed or expected to be employed at different time periods to achieve each level of performance. As was discussed above, envelopes representing the best of expected technology performance at different times are more reliable as predictors than are attempts to describe what any individual technology option will achieve. This is particularly true for technologies that have been advancing exponentially for many decades, as is the case for digital technology. In the absence of contradictory evidence, the panel's best projection for the future is continued exponential advance, recognizing that the advance will likely be stepwise rather than continuous. There is no reason to project lower or higher rates of growth for the future, at least until some fundamental technological barrier can be identified.
Digital technology is best characterized by the minimum achievable fabrication linewidth, the maximum area of defect-free chip that can be economically produced, and the clock speed achievable. The historical growth of minimum manufacturable mask fabrication linewidth projected to 2035 is illustrated in Figure 4.6. In many ways, progress in achievable linewidth explains much of the growth in devices per chip and clock speed—reducing the linewidth definition permits smaller devices to be fabricated, thereby increasing the number of devices we can implement on the same size chip while also encouraging faster switching. Additional benefits accrue from lower voltages and lower power per transistor as the dimensions shrink.
FIGURE 4.6 Microelectronic fabrication linewidth definition. SOURCE: Adapted from Yu, Albert, 1996, "The Future of Microprocessors," IEEE Micro, 16(6):46-53, Figure 3, December.
To date, the reduction in minimum linewidth definition shown in Figure 4.6 has been accomplished largely through continuous improvements in optical lithography-an achievement that was totally unexpected 10 years ago and that illustrates the difficulty of successfully projecting the growth of any specific technology. Although more expensive but finer-resolution x-ray and e-beam technologies have long waited in the wings, optical lithography has continued to maintain its dominance through shifts to shorter wavelengths and the introduction of subtle phase manipulation concepts.
This surprising result illustrates another difficult facet of projecting the details of technology growth. Anticipating a revolutionary new approach is clearly extremely difficult and very unlikely. But what is perhaps even more surprising is that the ability of an existing technology to improve significantly is often completely underestimated and in many ways just as difficult to predict. Given that the suppliers of the incumbent technology generally have much to lose and a fair amount of resources to apply in sharp contrast to the developers of revolutionary advances, it should not come as a surprise that the incumbent more often than not can maintain the lead.
Although data exist for the individual trends, the history of microprocessor clock speed, shown in Figure 4.7, provides an excellent summary of the overall growth of digital capabilities. A simultaneous measure of intrinsic switching speed as well as of the ability to manufacture useful complex circuits, microprocessor clock speed has been increasing at the rate of a factor of two every 2 years since the early 1980s. There is no evidence to suggest that this trend will not continue.
FIGURE 4.7 Clock frequencies of affordable microprocessors. SOURCE: Adapted from Hospodor, A.D., and A.S. Hoagland, 1993, "The Changing Nature of Disk Controllers," Proceedings of the IEEE, 81(4):586-594, Figure 10.
By 2005, the clock speeds of affordable (desktop) microprocessors are predicted to exceed 1 GHz—which is not surprising given that PCs, with 100- to 200-MHz clocks are already common. No single approach can accommodate all this growth, yet the specific technology that may achieve the best performance at any particular time is not predictable with confidence. Today's silicon circuitry will not move significantly into the gigahertz range. New technology will appear in the form of new, more capable device concepts, such as quantum dots and wires, and/or new material systems. GaAs is already capable of 10-GHz rates, and SiGe promises similar performance, in an easier-to-fabricate material system. InP devices could possibly operate at 100 GHz, which is anticipated near 2030. Well before this, superconducting RSFQ logic, with its 100- to 300-GHz potential, should become available in HTS technology. By 2035, the end of the time horizon of this study, currently envisioned RSFQ technology may approach fundamental limits and the technology of choice may have shifted to superconductor quantum dots or a family of devices yet to be invented.
Anticipating technology limits is generally a very difficult task, particularly if the deficiencies are of a practical rather than a theoretical nature. Linewidth, in contrast to other significant technology parameters, may encounter some fundamental obstacles to continued exponential improvement. The most obvious is that, although some day it might be possible to manipulate single atoms on the surface of a wafer, it seems to make no sense at all to discuss dimensions smaller than an atom or molecule. Checking the linewidth dimensions expected by 2035, one finds from Figure 4.6 an estimated fabrication capability of 1/5 to 1 nm, which corresponds to only a few interatomic distances for a silicon surface. Because it is not expected that linewidth fabrication technology will reach this absolute limit by 2035, some as yet unknown sophisticated techniques will have to be developed to permit manufacturing at these levels.
New Physics-Quantum Devices and Nanoelectronics
New types of elementary devices15 based on different physics may become available before atomic dimensional limits are encountered. The exploitation of quantum resonances in one dimension has already found its way into practice through the HEMT or modulation-doped field-effect transistors (MODFETs) and single and multiple quantum well (SQW and MQW) semiconductor lasers.
Precise control of thin-layer deposition to dimensions as small as 5 or 10 nm has been common in semiconductor manufacturing for years. The resulting one-dimensional resonances result in quantum confinement in the vertical dimension, leading to two-dimensional electron gases with a number of useful attributes, such as energy level splitting with restricted energy transfer between levels and a constant density of states.
With the ability to fabricate horizontal structures lithographically with dimensions of 50 to 100 nm, it becomes possible to provide quantum confinement in any or all three dimensions, leading to a variety of quantum plane, wire, and dot configurations. Because of the quantum effects, the number of electrons in the well of a quantum dot is quantized to an integer number of electrons, and even if the number is as large as tens to hundreds, single-electron changes can be observed.16 Combining quantum dots with very thin insulating layers, single-electron tunneling transistors have been implemented, and the possibility of single-electron logic is on the horizon. Although ingenious designs are being proposed, the less-than-unity voltage gain that characterizes the devices tested to date remains a fundamental unsolved engineering problem.
Research in this promising discipline has been active since its formulation in the mid-1980s. Even quantum molecules—i.e., several closely spaced quantum dots interacting via quantum mechanical tunneling-have been explored. All these single-electron semiconductor concepts apply equally well to superconductor implementations with a single Cooper pair replacing the single electron. The work to date has provided a sound theoretical and experimental basis for the underlying physics. The challenge remaining is to come up with practical, working, manufacturable devices in time to keep up with the exponential growth of digital electronics projected for the time frame beyond 2020. It seems likely that the transition from microelectronics to nanoelectronics and its future high-speed devices will depend on single-charge tunneling effects and the single-electron transistor. The Department of the Navy should pay close attention to this critical technology.
Analog-to-Digital Conversion
Soon, for practically all sensor applications, the detected signals will be digitized as close to the physical world interface as possible. The benefits of a digital signal representation over analog are familiar—fewer, easily miniaturized components for lower size and cost, immunity to component drift, control of computation-induced errors, and great algorithm flexibility. The translation of these analog signals into digital form with adequate signal bandwidth and dynamic range is of critical importance.
Conventional high-performance ADCs are strictly limited by the timing jitter associated with the comparators. Figure 4.8 summarizes the performance of available and projected ADC devices. No conventional ADC performs significantly above the limit indicated for a jitter level on the order of 1 ps root-mean-square.
FIGURE 4.8 State-of-the-art analog-to-digital converters. Current development of Delta-Sigma (DS) devices will add up to 6 bits by the year 2000.
Analysis indicates that the so-called jitter limit can be succinctly expressed by a single constant value of the product of dynamic range times sampling frequency, that is (2(B+1) -1) fsample = 8 × 1011 = 1 ps jitter limit, representing formally the familiar practical tradeoff between these two parameters.
In recent years, although the region of available ADCs has been expanding systematically into the higher sampling rates through developments in GaAs and InP technology, progress across the jitter boundary has moved at a glacial pace-by some estimates, it has been taking about 6 years to gain a single bit. A different concept has been sought-one that avoids the jitter issue entirely, if possible.
Such an approach, the Delta-Sigma (DS), was already available but as yet unused for RF sample rates. This technology was developed first for audio applications in the 1970s and 1980s and is inherent in the digital audio that underlies CDs and the like today. As a result of this long history, the concepts are well understood today and have proven robust as predicted, permitting high-dynamic-range implementations with inexpensive, modest-to-low-quality components. Today, for the first time, these proven concepts are being extended from megahertz to gigahertz rates and promise to make digital RF systems practical.
In this approach, the signal is greatly oversampled at a rate that permits a valid 1-bit representation of the difference between successive samples, and the resulting digital representation is passed back through a simple digital-to-analog (D/A) conversion and subtracted from the incoming analog signal to produce a form of differentiation. The resulting filter loop structure is used to move the quantization noise away from the spectral regime that contains the signal information in such a way that when the signal and related quantization noise are passed through a band pass filter, the reduced quantization noise levels are equivalent to having digitized the signal directly in the conventional manner with multiple-bit accuracy. The output digital data stream is then reduced to the rates appropriate to the information bandwidth of the signal. Depending on the order of the loop structure, the gain in effective number of bits can be 1.5, 2.5, and 3.5 bits, and so on, per octave of oversampling for 1, 2, 3, and so on, pole filters. In strong contrast, the conventional approaches to multibit A/D, which do not try to shape the noise, achieve only a one-half bit per octave increase in effective bits with oversampling.
The loop averages out jitter effects in exchange for new problems—e.g., the higher the order of the loop, the more effective the bits per octave, but the more difficult it is to keep stable. The DS approach will take us rapidly beyond the jitter barrier of existing ADC for the larger information bandwidths and dynamic ranges, characteristic of high-performance sensors of interest in this study. Successful development of gigahertz DS technology will produce a distinct jump in ADC performance in the near future. In view of the projected increases in clock speeds, this technology can be expected to provide several bits, i.e., 3.5 per octave for a three-pole loop, every 2 years or so, as additional octaves of oversampling become available through the continuing growth of digital clock speed capabilities.
In Figure 4.8, the points labeled DS (Delta-Sigma) represent the performance targeted for several noncomparator ADC architectures currently under development for RF sampling rates. Included are one to be implementated in SiGe and one each to be implemented in high-temperature and low-temperature superconductor technology. Both superconductor efforts are addressing the same performance—20 equivalent bits at 20 MHz.
Packaging and Interconnections
Packaging and interconnections, often underestimated or even ignored, are of enormous importance. It does little good to develop powerful digital chips if they cannot be effectively and reliably packaged and interconnected. Some digital circuits already are limited in performance more by the packaging and signal interconnection characteristics than by the basic clock speeds of the individual chips. Performance and packaging are interdependent, and in recent years these interdependencies have grown so strong that it is becoming increasingly difficult to separate the devices from their packaging. Future generations of sensors will utilize very-high-speed, large-area, extremely complex logic and memory chips that must be interconnected and packaged within severe constraints on volume, weight, power-handling capability, and system operating speeds. The digital electronics will be completely integrated into a single unit, and its design and physical implementation, including packaging, will have to be approached from the point of view of a single entity.
Although monolithic or full wafer-scale integration has been a dream for many years, current practice targets the more attainable goal of hybrid wafer-scale integration. In the MCM approach, bare chips—that is, IC chips without packaging of any kind—are attached to a wafer or wafer-like substrate that contains the passive circuitry to supply the chip-to-chip interconnections, very similar to a small backplane or printed wiring board. Starting with two-dimensional configurations, MCM technology now includes consideration of three-dimensional stacked structures that, although providing the desired small sizes, bring up serious challenges in heat removal, z-axis signal interconnectivity, and test and repair. Since such complex modules are both expensive and very difficult to access internally once fabricated, the sensor architecture should be designed with fault-tolerant elements providing resource redundancies and autonomous reconfiguration capabilities.
Many MCM options have been developed, utilizing a wide variety of substrate materials, chip-to-substrate attachment approaches, electrical-signal interconnect technologies—both in-plane, from chip-to-chip, and vertically, from substrate-to-substrate—and thermal control techniques. As yet, no single MCM approach clearly dominates, that is, is optimum for all applications. It is too early to tell if future packaging will evolve as a refinement of the MCM hybrid concepts or whether fabrication practices will eventually permit monolithic, wafer-scale implementations, or whether something else entirely unexpected will appear.
As clock speeds increase, moving signals from place to place without significant loss or distortion becomes increasingly difficult. Currently, within computers or other digital processors, high-speed signals are routed electrically via controlled impedance lines with impedance-matched connections. Only between widely separated computers are high-bandwidth modern fiber-optic communication links routinely employed. Optical interconnects exhibit much less signal loss with propagation distance than does an electrical interconnect and, because of the high frequency of the optical carrier, easily support very-large-bandwidth signals. However, there are serious efficiency penalties associated with modulating the electrical signal information onto and off the optical carrier that are not present in the direct electrical approaches.
In the future, as clock speeds and information bandwidths continue to grow, free-space and fiber-guided optical interconnects will certainly become the preferred way to transfer the ever-growing number of high-speed signals from board to board or even chip to chip.
Whatever the details, it is extremely important that packaging and interconnection considerations so critical to the success of all future high-performance sensors be addressed and be factored into chip design from the beginning, as part of the overall architecture. These technical factors have to be an integral and explicit part of the Navy's future R&D plans.
Computer Technology
Little need be said here about the future of computer technology and its impact on future sensor capabilities. The exponential growth of computer capabilities can be expected to continue throughout the many decades of interest to this study, fueled by advances in device technology, architectures, and software. As the systems grow in the direction of ever larger distributed networks of interconnected multiprocessors, the implications for reliability must be carefully taken into account and both fail-safe and fault-tolerant mechanisms included so that a single failed component cannot pull down the whole network or completely invalidate the outputs.
A history of the growth in throughput capabilities for both supercomputers and affordable (desktop) computers is described in more detail in Chapter 2 of this report.
Perhaps the most interesting possibilities lie not so much with the large, multicomponent computers certain to be needed for powerful sensors systems, such as SAR radars and sonars, digital adaptive beamforming phased arrays, multisensor surveillance networks exploiting data fusion, and so on, but rather with the potential for combining, or in the extreme, integrating, monolithically on the same substrate, sensors with an ADC, a digital processor, and a communication output port to make a complete sensor system on a chip.
Inevitably, multiple copies of these smart sensors will be configured into intercommunicating networks that will act together to perform as a single composite metasensor, sharing, combining, and fusing the individual sensor's data into a total sensor data view that is more powerful than just the sum of the parts. Adding mobility to the individual sensors, useful ant-like societies of intercommunicating microrobot sensors, functioning as large distributed, spatially adaptive, extraordinarily capable surveillance systems, no doubt will follow.
Algorithm Technology
To exploit the enormous opportunities that the exponential growth of digital computing capabilities will continue to provide in the future requires algorithms. For some sensor applications, many years of effort have already been applied and effective algorithms are known to exist. Some of these algorithms find application today, but others have not yet been introduced into practice because the computer or digital signal-processing support available for operational deployment is inadequate to permit real-time implementation. That is, the real-time configurations permitted by today's technology are too large, require too much power, or are simply too expensive for the intended application. The future will certainly change this picture dramatically.
Adaptive Processing
As available computational resources grow beyond the simple requirements of the sensor, whatever that may be—radar, sonar, and electro-optical—it is natural to apply the additional computational power to improving sensor performance by continuously optimizing its processing in response to the observed data. Several easily implemented analog adaptive techniques have long been employed in radar and sonar-constant false alarm (CFAR) detection, with an adjustable detection threshold to prevent the processing of false alarms from overwhelming the real-time capability of the system, and sidelobe cancellation utilizing an adjunct omnidirectional antenna, to distinguish between main beam and sidelobe returns.
With enough digital-computing resources, much more progress can be envisioned. Interfering signals, which are localized in space or frequency or both, can be estimated from the measurements and explicitly subtracted or canceled to improve the detection of the desired target signals. For array antenna systems, performance in the presence of jammers can be greatly enhanced by estimating the direction of the interfering sources and choosing the weights to be applied to the signals from each element such that a null is placed in the antenna pattern in the direction of the interfering signal. For computational reasons, to date these ideas have been applied only to small antennas, i.e., a few to a few tens of elements, or to very-low-frequency radars with several hundred elements—the Navy's operational, relocatable over-the-horizon radar (ROTHR) uses digital beamforming, for example. As it will soon be practical to digitize the return signals on each receive antenna element close to, or at the front end of, a radar, even at X-band and above, digital beamforming of large arrays with hundreds to thousands of elements can be expected in the next 5 to 10 years for a wide range of applications. These possibilities will be developed further under digital radars.
For clutter suppression, similar concepts apply, with the exception that although jammers are typically spectrally complex but well localized spatially, i.e., point sources, clutter is distributed in both space and spectrum. In the particular case of airborne radar, the ground-clutter returns have Doppler offsets that are determined by the aircraft velocity and by the angle between the line of sight to the patch of ground and the aircraft's velocity vector. The solution is thus to introduce spatial nulls into the received beam patterns separately for each Doppler bin. This approach, known as space-time adaptive processing (STAP), is obviously computationally demanding and has yet to be applied to any operational system but is certainly a candidate for the near-term Joint Strike Fighter (JSF) generation of airborne radars.
One of the most intriguing aspects of digital adaptive beamforming is the potential for enabling the use of less-than-perfect radar antennas or optical configurations that can be digitally corrected to provide near-optimal system performance. In this way, digital processing can be used to alleviate difficult manufacturing issues (associated with high-precision optical components) or to permit antennas to be optimized for other purposes, such as reduced signature.
Automatic Target Recognition
As the volume of sensor-provided data increases and the operational time of advanced weapon systems decreases, automatic target recognition (ATR) becomes mandatory—no human can provide the necessary decisions fast enough. Although certainly challenging, ATR has acquired a reputation for impossibility that does not reflect the facts. It is the case that, given two- and three-dimensional image data visually displayed, and enough time, a trained human is unbeatable in many recognition tasks. Considering for example, the task of finding a familiar face in the crowd without knowing in advance who it is going to be: A human can do this if the face is familiar to the observer; a computer cannot. But this does not mean that computer-based ATR is impossible. And it does not mean that ATR will not outperform the human in some circumstances.
Several effective ATR applications have already been demonstrated—search-and-destroy armor munition (SADARM) recognizes its specific target automatically through millimeter-wave passive imagery, brilliant antitank (BAT) munition uses a combination of acoustic and IR imagery, and the tactical LIDAR seeker uses LIDAR three-dimensional range-to-pixel imagery to automatically detect and recognize a variety of military targets, such as tanks, trucks, and bridges. In each case a simple form of ATR, tuned to the characteristics of the scenario and simplified to permit real-time implementation, was employed.
Increasing the sophistication and effectiveness of the ATR algorithms faces several obstacles. These challenges for the future include the following:
Fast Algorithms
Whatever the algorithm, the fewer the operations required to implement it, the fewer are the computer resources required and the more practical the algorithm becomes. Many effective algorithms-singular value decomposition (SVD) is a good example—are languishing today because the computer resources have not yet met their requirements. The development of a fast algorithm can change everything. The best known example of a fast algorithm is the fast Fourier transform (FFT) that, by exploiting the many symmetries of the discrete Fourier transform, reduced computational requirements without affecting the accuracy of the results. Of course, the growth of computer capabilities will eventually enable brute force implementation of any algorithm, but a fast form will always offer advantages by freeing up computer resources for other tasks. The creation of fast algorithms should remain a focus of research during the period from 2000 to 2035 and beyond.
New Algorithms
From time to time completely new approaches to signal or information processing arise and should be watched and carefully assessed for potential application to sensors. Some such advances represent the development of new mathematics, such as wavelets and chaos, which have given rise to interesting concepts in multirate and nonlinear signal analysis with potential for superior sensor performance. Other advances can arise from attempting to understand the superior performance of some biological systems, such as the sonars of bats or dolphins or the navigational abilities of birds or sea turtles. Enhanced sonar performance already has been demonstrated using biologically motivated (e.g., bat) signal processing.
Information and Data Extraction
If sensor technology progresses as envisioned, with multiple, multispectral sensor systems and increasing rates of data sampling, the amount of data collected could easily overwhelm the user. In many ways, too much data is equivalent to no data at all. In either case, no useful information is forthcoming. Although ATR goes a way toward resolving this dilemma through the processing of raw data into compressed information—e.g., target type Q, detected at position x-y, with parameters {a, b, c, ..}, and so on—not all sensor-produced information is a simple list of targets. It may be a picture of what the sensor is seeing, yet what is important differs from one user to another. The challenge for the future is to develop information-identification and -extraction algorithms that can identify and encode the interesting information in the data, whatever that might be, so that redundant or uninteresting portions of the data may be deleted or so that the full data set may be characterized such that only those subsets that contain a particular form of information need to be examined in full detail. Better means of identifying information contained in large sensor-generated databases are required, as are better search strategies to efficiently exploit this information.
Data Compression
The final algorithm challenge lies in the necessity to communicate between sensors and from sensors to the users. The reality is that communication bandwidths, particularly those in an active battle zone, will always be severely limited and certainly not capable of moving all of the raw data generated by all the sensors everywhere. The data must be compressed for efficient communication of the significant information without losing important portions or introducing artifacts or false alarms. A major portion of the solution to this communication problem lies in the development of information- and data-extraction algorithms as discussed above.
More straightforward are the more mechanical aspects of encoding the data to be transmitted once the critical information has been extracted in minimal form, although there remain some unresolved issues. If the data are to be transmitted in completely uncorrupted form, it can only be compressed by small factors of two or three by lossless encoding, and this gain may be seriously compromised by the need to apply heavy transmission redundancy encoding if the channel is unreliable, i.e., noisy or jammed. For this situation, the real gain is going to be in information- and data-extraction algorithms.
On the other hand, if imagery to be examined by the user is to be sent, lossy compression-sometimes by factors of 40 or 50 to 1—may be acceptable. Experiments with medical imagery at these levels of lossy compression have shown little or no difference from the original images in the diagnostics resulting from these images. For two-dimensional images, efficient and effective compression schemes, which exploit the local continuity of the images, have been developed using various transform techniques, including Fourier transforms and wavelets. For higher-order images, that is, higher-dimensional data sets, such algorithms need to be formulated and proven. This is another important and promising area of research for the future.
Technology Vulnerabilities
For many of the technologies just discussed, improved performance carries with it the potential for enhanced or additional vulnerabilities—to deliberate countermeasures or perhaps simply to the environment. Table 4.2 summarizes the most obvious ones. The sources of most of these weaknesses lie with the twofold trends toward low-voltage electronic implementations and complexity.
| Technology | Vulnerabilities |
| Low-voltage devices | Electromagnetic interference susceptibilities |
| Superconductors | Cryo dependencies |
| Very-low-voltage circuits, EMI | |
| Computers | Intrinsic reliability, fault tolerance |
| Software | Intrinsic reliability |
| Latent faults, correctness | |
| Viruses, time bombs | |
| Malignant applets or beans | |
| Nanotechnology | Tasking, networking, fail-safe operation |
From the device point of view, as fabrication linewidths decrease, operating voltages and signal levels also decrease in order to avoid excess electric-field conditions, making the electronics more sensitive to noise and pickup effects also known as electromagnetic interference (EMI). This suggests that future, high-performance smart sensors may be quite vulnerable to weapons that employ EMP generation, unless careful consideration is given from the start to providing inherent shielding as part of the packaging. Similar issues arise in semiconductor systems that exploit superconductor bandgap phenomena, such as SQUIDs and RSFQ logic devices, and as a result use voltages comparable to the bandgap, i.e., millivolts.
From the system side, as individual digital devices grow smaller and cheaper and simultaneously more computationally capable, the systems grow more complex. With increasing clock speeds, storing and distributing the data become increasingly challenging. Multiprocessor configurations introduce new issues of coordination and reliability—the systems become increasingly opaque, i.e., harder to understand and to confirm or guarantee correctness, while simultaneously becoming easier to upset by malicious actions. The cooperating societies of sensors envisioned for the future involve even higher levels of complexity and additional communications vulnerabilities, and also raise difficult questions of sensor autonomy, fail-safe behavior, and the potential for insanity—e.g., could a sensor fail, or be induced to fail, to the point of turning a weapon against its owners?
These generic technology vulnerabilities must be addressed continuously during the development of advanced sensor systems and included explicitly in the assessments of the proposed sensor's utility.
INDIVIDUAL SENSORS
Having discussed the state of the art and
the observed growth patterns of the five technologies identified as generic
to all sensors, the panel now describes the advanced sensors known or thought
to be critical to future naval applications that these technologies enable.
Each of the individual sensor classes summarized in Box 4.5 is discussed
briefly below, with the implications of the general technology trends and
projected growth factored into the context of each sensor class.
| BOX 4.5 Classes of Individual
Sensors |
Electromagnetic Sensors
Radar
Radar, with its all-weather, long-range capabilities for detection and tracking, is the primary electromagnetic sensor in the Navy's tool box and promises to be useful throughout the time horizon of this study. Most of the common technology trends identified earlier are exhibited by radar technology and provide excellent guidance to the potential future capabilities and applications of this key sensor class.
Phased Arrays
For many decades, radars have been evolving toward distributed phased-array configurations. Given the proven flexible, rapid beamsteering capabilities of electronic scan, with few exceptions new radar implementations—whether surface based or airborne—are envisioned to be phased arrays. Complementing this transition to phased arrays has been the steady replacement of classical microwave tubes and waveguide-based analog components by solid-state microwave equivalents—oscillators, low-noise and low-power amplifiers, phase shifters, complete T/R modules, and the like—in MMIC technology.
Monolithic Microwave Integrated Circuit Technology
For the past decade and a half, as the technology matured, MMIC-based T/R modules have increased steadily in performance while declining in cost, albeit more slowly than desired. With GaAs technology, the output power per module has grown exponentially in the common radar bands, while simultaneous exploitation of decreasing fabrication linewidths combined with new material systems, such as InP and soon SiGe, has extended the operating frequencies through the millimeter-wave bands to 94 GHz and higher. Continuing exponential increases in the performance of individual devices are expected as wide-bandgap material technologies (e.g., SiC, GaN, and diamond) mature.
Millimeter Wave Frequency
Operating at millimeter-wave frequencies, however, brings with it more serious obstacles to phased-array implementations than simple power generation. At 94 GHz, a millimeter-wave frequency that coincides with a spectral region of minimal atmospheric absorption, the wavelength is only 3 mm, and lower-frequency discrete structural-assembly techniques for half-wavelength separation become difficult, if not impossible, to implement in any straightforward manner. New concepts for continuously distributed phased-array implementations become necessary—calling for higher levels of integration for which the distinction between the array elements and the packaging ultimately should vanish. In spite of the inherent implementation difficulties, precisely because of the small wavelengths, millimeter-wave multibeam antennas offer the potential for generating high-resolution two-dimensional imagery from conveniently small physical antenna. The use of passive millimeter-wave imaging systems offers interesting, perhaps even breakthrough, possibilities for foliage- and camouflage-penetration.
Module Costs
The overall affordability of a phased-array radar is particularly sensitive to the cost of an individual T/R module. To suppress the large spurious sidelobes known as grating lobes, phased-array elements are usually distributed uniformly across the face of the antenna with an element-to-element separation of not more than a half wavelength. On the other hand, good transverse resolution often requires aperture dimensions that are hundreds of wavelengths across. The result of these two conditions is arrays with hundreds to thousands of array elements. Unless the cost of each element is relatively small, phased arrays may simply not be affordable for many applications. For example, today the cost per element (i.e., total antenna structure cost divided by the number of T/R modules) of an X-band T/R module similar to what is used in the Army's ground-based radar is between $1,000 and $2,000, and so the total cost of such a phased-array antenna (~ 20,000 elements) will be millions of dollars. Current efforts target reduction in production costs of an order of magnitude through a combination of increased integration and the manufacturing learning curve. A $300 T/R module has long been the goal, but the current high module costs inhibit the purchase of large, phased-array radars, which in turn reduces the number of T/R modules to be manufactured, thereby limiting the manufacturing learning experience-a chicken and egg situation. There seems little doubt, however, that this situation will eventually be resolved in favor of solid-state phased-array technology and costs will gradually fall into an acceptable,