Emerging Challenges: Mobile Networking

for “Smart Dust”

Joseph M. Kahn, Randy Howard Katz, and Kristofer S. J. Pister

Abstract: Large-scale networks of wireless sensors are becoming

increasingly tractable. Advances in hardware technology and engi-
neering design have led to dramatic reductions in size, power con-
sumption and cost for digital circuitry, wireless communications
and Micro ElectroMechanical Systems (MEMS). This has enabled
very compact, autonomous and mobile nodes, each containing one
or more sensors, computation and communication capabilities, and
a power supply. The missing ingredient is the networking and ap-
plications layers needed to harness this revolutionary capability
into a complete system. We review the key elements of the emer- Fig. 1. Cubic-inch wireless sensor nodes (motes) constructed using
gent technology of “Smart Dust” and outline the research chal- off-the-shelf technology. On the left is a radio-frequency mote with
temperature and light sensors. On the right is a laser mote with tem-
lenges they present to the mobile networking and systems commu- perature, light, humidity, barometric pressure, and sensors.
nity, which must provide coherent connectivity to large numbers of
mobile network nodes co-located within a small volume.
With improvements in integration, packaging, circuit design,
and process technology, autonomous sensor nodes like these
I. INTRODUCTION will continue to shrink in size and power consumption while
As the research community searches for the processing plat- growing in capability. Berkeley’s Smart Dust project, led by
form beyond the personal computer, networks of wireless sen- Professors Pister and Kahn, explores the limits on size and
sors have become quite interesting as a new environment in power consumption in such autonomous sensor nodes.
which to seek research challenges. Many researchers have re- Size reduction is paramount, to make the nodes as inexpen-
cently shown that it is possible to integrate sensing, communi- sive and easy-to-deploy as possible. The research team is work-
cation, and power supply into an inch-scale device using only ing to incorporate the requisite sensing, communication, and
off-the-shelf technology. These have been enabled by the rapid computing hardware, along with a power supply, in a volume no
convergence of three key technologies: digital circuitry, wire- more than a cubic millimeter, while still achieving impressive
less communications, and Micro ElectroMechanical Systems performance in terms of sensor functionality and communica-
(MEMS). In each area, advances in hardware technology and tions capability. These millimeter-scale nodes are called “Smart
engineering design have led to reductions in size, power con- Dust.”Although mimicking the mobility of dust is not a primary
sumption, and cost. This has enabled remarkably compact, au- goal, future prototypes of Smart Dust will be small enough to
tonomous nodes, each containing one or more sensors, compu- remain suspended in air, buoyed by air currents, sensing and
tation and communication capabilities, and a power supply. communicating for hours or days on end. At least one popular
Fig. 1 shows two examples of this off-the-shelf sensor tech- science fiction author has articulated just such a vision [1].
nology. The device on the left contains a microprocessor, tem- In this paper, we are concerned with the networking and ap-
perature sensor, light sensor, 900 MHz radio, and battery (hid- plications challenges presented by this radical new technology.
den underneath). The radio range is about 10 m. The device on These kinds of networking nodes must consume extremely low
the right consists of a microprocessor and four sensors, and the power, communicate at average bit rates measured in kilobits
radio has been replaced with a laser pointer driven by the micro- per second, and potentially need to operate in high volumetric
processor. Transmitting at 4 bps to a small CCD camera attached densities. These requirements dictate the need for novel ad hoc
to a PCMCIA frame grabber in a laptop computer, that device routing and media access solutions. Smart dust will enable an
is capable of sensing and communicating weather information unusual range of applications, from sensor-rich “Smart spaces”
at a distance of over 20 km. Communication was demonstrated to self-identification and history tracking for virtually any kind
from San Francisco to the authors’ building in Berkeley, across of physical object.
the San Francisco Bay. Both devices have a full-duty lifetime of The study of “Smart Dust systems” is very new. The main
about a day. purpose of this paper is to present some of the technologi-
cal opportunities and challenges, with the goal of getting more
Manuscript received June 2000. systems-level researchers interested in this critical area. The re-
The authors are with the department of Electrical Engineering and
Computer Sciences, University of California, Berkeley, USA, e-mail: mainder of this paper is organized as follows. Section II presents
jmk@eecs.berkeley.edu, randy@eecs.berkeley.edu, pister@eecs.berkeley.edu. an overview of the technology that underlies Smart Dust. Sec-

1229-2370/00/$10.00 c 2000 KICS
Fig. 2. Smart dust mote, containing microfabricated sensors, optical receiver, passive and active optical transmitters, signal-processing and control
circuitry, and power sources.

tion III outlines the key networking challenges presented by this much energy as possible when the sun shines (roughly 1 J per
technology. In Section IV, we describe some of the potential day) or when room lights are turned on (about 1 mJ per day).
applications of Smart Dust and the challenges they pose. Sec- With energy as the most precious resource, and time not as
tion V discusses related projects from the research community. likely to be critical, elementary operations and ultimately algo-
Section VI presents our summary and conclusions. rithms are likely to be judged in terms of their energy cost, rather
than power consumption. Energy-optimized microprocessors
currently use roughly 1 nJ per 32-bit instruction. Commer-
II. SMART DUST TECHNOLOGY cially available data acquisition chips approach 1 nJ per sample.
The goal of the project, a Smart Dust mote is illustrated in Bluetooth radio-frequency (RF) communication chips will burn
Fig. 2. Integrated into a single package are MEMS sensors, a about 100 nJ per bit transmitted. There is much excitement in
semiconductor laser diode and MEMS beam-steering mirror for the RF circuit community right now over picoradios, which tar-
active optical transmission, a MEMS corner-cube retroreflec- get 1 nJ/bit. Dramatic improvements will be made in some of
tor for passive optical transmission, an optical receiver, signal- these categories, but for every Joule that a mote stores, it will
processing and control circuitry, and a power source based on have the ability to perform some billions of operations. The net-
thick-film batteries and solar cells. This remarkable package working and information theory challenge is to determine how
will have the ability to sense and communicate, and be self- to allocate the energy: Sense? Compute? Transmit?
powered! Techniques for performing sensing and computation at low
A major challenge is to incorporate all these functions while power are reasonably well understood. Developing a communi-
maintaining very low power consumption, thereby maximizing cations architecture for ultra-low-power represents a more criti-
operating life given the limited volume available for energy stor- cal challenge. The primary candidate communication technolo-
age. Within the design goal of a cubic millimeter volume, using gies are based on RF or optical transmission techniques. Each
the best available battery technology, the total stored energy is technique has its advantages and disadvantages. RF presents a
on the order of 1 J. If this energy is consumed continuously over problem because dust motes offer very limited space for anten-
a day, the dust mote power consumption cannot exceed roughly nas, thereby demanding extremely short-wavelength (i.e., high-
10 W. For comparison, this is roughly the “Shutdown” power frequency) transmission. Communication in this regime is not
of individual low power ICs found in today’s laptop computers. currently compatible with low power operation. Furthermore,
The functionality envisioned for Smart Dust can be achieved radio transceivers are relatively complex circuits, making it dif-
only if the total power consumption of a dust mote is limited ficult to reduce their power consumption to the required mi-
to microwatt levels, and if careful power management strategies crowatt levels. They require modulation, bandpass filtering and
are utilized (i.e., the various parts of the dust mote are powered demodulation circuitry, and additional circuitry is required if the
on only when necessary). To enable dust motes to function over transmissions of a large number of dust motes are to be multi-
the span of days, solar cells could be employed to scavenge as plexed using time-, frequency- or code-division multiple access
[2].
An attractive alternative is to employ free-space optical trans-
mission. Studies have shown that when a line-of-sight path is
available, well-designed free-space optical links require signifi-
cantly lower energy per bit than their RF counterparts [2]. There
are several reasons for the power advantage of optical links. Op-
tical transceivers require only simple baseband analog and digi-
tal circuitry; no modulators, active bandpass filters or demodula-
tors are needed. The short wavelength of visible or near-infrared
light (of the order of 1 m) makes it possible for a millimeter-
scale device to emit a narrow beam (i.e., high antenna gain can
be achieved). As another consequence of this short wavelength,
a base-station transceiver (BTS) equipped with a compact imag-
ing receiver can decode the simultaneous transmissions from a
large number of dust motes at different locations within the re-
ceiver field of view, which is a form of space-division multiplex-
ing. Fig. 3. Microfabricated corner-cube retroreflector, consisting of three
Successful decoding of these simultaneous transmissions re- gold-coated polysilicon mirrors. The base mirror can be deflected
electrostatically, modulating the optical signal reflected from the de-
quires that dust motes not block one another’s line of sight to the vice (taken from [3]).
BTS. Such blockage is unlikely, in view of the dust motes’ small
size. A second requirement for decoding of simultaneous trans-
mission is that the images of different dust motes be formed on has important implications for feasible network routing strate-
different pixels in the BTS imaging receiver. To get a feeling for gies (see Section III-A.2).
the required receiver resolution, consider the following exam- Fig. 4 illustrates a free-space optical network utilizing the
ple. Suppose that the BTS views a 17 m  17 m area containing CCR-based passive uplink. The BTS contains a laser whose
Smart Dust, and that it uses a high-speed video camera with a beam illuminates an area containing dust motes. This beam can
256  256 pixel imaging array. Each pixel views an area about be modulated with downlink data, including commands to wake
6.6 cm2 . Hence, simultaneous transmissions can be decoded as up and query the dust motes. When the illuminating beam is not
long as the dust motes are separated by a distance roughly the modulated, the dust motes can use their CCRs to transmit up-
size of a pack of cigarettes. link data back to the base station. A high-frame-rate CCD video
Another advantage of free-space optical transmission is that a camera at the BTS “sees” these CCR signals as lights blink-
special MEMS structure makes it possible for dust motes to use ing on and off. It decodes these blinking images to yield the
passive optical transmission techniques, i.e., to transmit modu- uplink data. Kahn and Pister’s analysis show that this uplink
lated optical signals without supplying any optical power. This scheme achieves several kilobits per second over hundreds of
structure is a corner-cube retroreflector, or CCR (see Fig. 3). It meters in full sunlight [2]. At night, in clear, still air, the range
comprises three mutually perpendicular mirrors of gold-coated should extend to at least a kilometer. Because the camera uses
polysilicon. The CCR has the property that any incident ray of an imaging process to separate the simultaneous transmissions
light is reflected back to the source (provided that it is incident from dust motes at different locations, we say that it uses space-
within a certain range of angles centered about the cube’s body division multiplexing. The ability for a video camera to resolve
diagonal). If one of the mirrors is misaligned, this retroreflec- these transmissions is a consequence of the short wavelength of
tion property is spoiled. The microfabricated CCR includes an visible or near-infrared light. This does not require any coordi-
electrostatic actuator that can deflect one of the mirrors at kilo- nation among the dust motes, and thus, it does not complicate
hertz rates. It has been demonstrated that a CCR illuminated their design.
by an external light source can transmit back a modulated sig- When the application requires dust motes to use active op-
nal at kilobits per second. Since the dust mote itself does not tical transmitters, MEMS technology can be used to assemble
emit light, the passive transmitter consumes little power. Us- a semiconductor laser, a collimating lens and a beam-steering
ing a microfabricated CCR, Chu et al. have demonstrated data micro-mirror, as shown in Fig. 2. Active transmitters make pos-
transmission at a bit rate up to 1 kbps, and over a range up to sible peer-to-peer communication between dust motes, provided
150 m, using a 5-mW illuminating laser [3]. there exists a line-of-sight path between them. Power consump-
It should be emphasized that CCR-based passive optical links tion imposes a trade-off between bandwidth and range. The dust
require an uninterrupted line-of-sight path. Moreover, a CCR- motes can communicate over longer ranges (tens of kilometers)
based passive transmitter is inherently directional; a CCR can at low data rates or higher bit rates (megabits per second) over
transmit to the BTS only when the CCR body diagonal happens shorter distances. The relatively high power consumption of
to point directly toward the BTS, within a few tens of degrees. semiconductor lasers (of the order of 1 mW) dictates that these
A passive transmitter can be made more omnidirectional by em- active transmitters be used for short-duration burst-mode com-
ploying several CCRs oriented in different directions, at the ex- munication only. Sensor networks using active dust mote trans-
pense of increased dust mote size. If a dust mote employs only mitters will require some protocol for dust motes to aim their
one or a few CCRs, the lack of omnidirectional transmission beams toward the receiving parties.
Fig. 4. Design of a free-space optical network in which a base-station transceiver communicates simultaneouly with a collection of many dust motes
(only one dust mote is shown). A single laser at the base station supplies optical power for the downlink and the uplink.

III. MOBILE NETWORKING CHALLENGES is more likely when the dust mote density is higher. Multihop
A. Overview routing increases latency, and requires dust motes to be equipped
with active optical transmitters. Constraints on size and power
Development of mobile networking protocols for Smart Dust consumption of the dust mote digital circuitry dictate the need
represents a significant challenge. Some critical limitations are: for low-complexity ad hoc multihop routing algorithms.
1) The free-space optical links requires uninterrupted line-of- When dust motes are floating in the air or otherwise not fixed,
sight paths, 2) the passive and active dust mote transmitters have a line-of-sight path to the BTS may become intermittently avail-
directional characteristics that must be considered in system de- able. In such cases, the BTS can continuously interrogate the
sign, and 3) there are severe trade-offs between bit rate, energy dust motes. When a line-of-sight path to a mote becomes avail-
per bit, distance and directionality in these energy-limited free- able, the mote can transmit a packet to the BTS. When the av-
space optical links. These limitations are described in more de- erage time between occurrence of viable line-of-sight paths is
tail in the following subsections. much longer than the packet duration, latency will probably be
minimized by using multihop routing instead.
A.1 Line-of-Sight Requirement
An unbroken line-of-sight path is normally required for oper- A.2 Link Directionality
ation of free-space optical links for Smart Dust. These links can-
not operate reliably using non-line-of-sight propagation, which In most Smart Dust systems, the BTS interrogating beam an-
would rely on reflections from one or more objects between gular spread should be matched to the field of view of the BTS
the transmitter and receiver. As shown in Section III-A.3, the imaging receiver. These two should be matched in all systems
transmitted beam should have a small angular spread in order to using passive dust mote transmitters, and in systems using active
achieve a high signal-to-noise ratio with acceptably small trans- dust mote transmitters when the application involves frequent
mitter power. Specular reflection may not significantly increase bi-directional transmission between the BTS and dust motes.
a beam’s angular spread, but the existence of a properly aligned Intuitively, it makes little sense for the BTS to interrogate dust
specular reflector would be a rare event. Diffuse reflection scat- motes from which it cannot receive, and vice versa. In these
ters a beam’s energy over a wide range of angles, making align- systems, the interrogating beam and imaging receiver will be
ment less critical, but usually scatters insufficient energy toward mounted rigidly together in the BTS, and will be aimed together
the receiver. Hence, diffuse, non-line-of-sight transmission is as a unit. For example, the BTS may reside in a hand-held unit
likely to be feasible only when active transmitters are used over resembling a pair of binoculars, which is aimed by a human op-
very short distances (probably under 1 meter). It is probably im- erator.
possible to use diffuse, non-line-of-sight transmission with pas- In certain applications using active dust mote transmitters, it
sive transmitters (based on CCRs), because both the interrogat- may be desirable to use a BTS transmitter beam whose angular
ing beam and the reflected beam would be subject to scattering spread is smaller than the BTS receiver field of view. In these
over a wide range of angles. applications, the interrogating beam will be aimed at various
A fixed dust mote without a line-of-sight path to the BTS can locations within the receiver field of view.
communicate with the BTS via multihop routing, provided that Because of limited available space, the dust mote’s optical re-
a suitable multihop path exists. The existence of such a path ceiver probably cannot employ an imaging or non-imaging op-
tical concentrator in front of the photodetector. As a result, the It is known that in free-space optical networks, non-
dust mote receiver will be fairly omnidirectional, i.e., it will be reciprocity can lead to “hidden nodes” which can cause col-
able to receive from most of the hemisphere located in front of lisions during medium access. For example, this effect is ob-
the dust mote. In most applications, it should not be necessary served in networks having a shared-bus physical topology, and
to aim the dust mote receiver. using MAC protocols based on random time-division multiplex-
The dust mote’s transmitter will exhibit markedly different ing, such as CSMA-CA with RTS/CTS [4]. In Smart Dust net-
directional characteristics than its receiver. A passive dust mote works, the uplink (dust mote to BTS) uses space-division mul-
transmitter is based on the CCR. This device reflects light di- tiplexing. As discussed in Section II, uplink collisions will not
rectly back to the source within a narrow beam 1 , provided that occur as long as the dust motes are sufficiently separated that
it is illuminated from a direction that lies within a few tens of their transmissions are detected by different pixels in the BTS
degrees of the cube body diagonal. If dust motes use only one imaging receiver. Collisions during active peer-to-peer commu-
CCR each, then any given dust mote, if fixed in a random, up- nications are a potential problem in Smart Dust networks. A
right orientation, has only about a 10% probability of being able peer-to-peer collision avoidance scheme must cope with a dy-
to transmit to the BTS. This probability can be increased signif- namic network configuration, while not introducing excessive
icantly by equipping each dust mote with several CCRs, each complexity or latency.
oriented along a different direction. As an alternative, a single
CCR may be mounted on a MEMS aiming mechanism. This A.3 Trade-Offs Between Bit Rate, Distance and Energy per Bit
mechanism need only aim the CCR with an accuracy of the or-
der of 10 or 20 
Free-space optical links are subject to trade-offs between sev-
eral design parameters. For simplicity, we consider the case of
Still other solutions exist for coping with the CCR’s direction-
links employing active laser transmitters. The receiver signal-
ality. It may be possible to distribute randomly an excess num- to-noise ratio (SNR) is given by
ber of dust motes, with the goal of communicating only with
SNR = C ENb Rd4bA4 :
those whose CCRs happen to point toward the BTS. If the dust 2 2
motes are not fixed, it may be best for a dust mote to simply
(1)
0
delay transmitting until it moves into an orientation that enables
transmission to the BTS. Here, C is a constant, Eb is the average transmitted energy per
An active dust mote transmitter is based on a laser diode. It bit, Rb is the bit rate, A is the receiver light collection area 2 , N0
should employ a narrow beamwidth, typically less than a degree is the receiver noise power spectral density, d is the link trans-
(see Section III-A.3). This necessitates equipping the dust mote mission distance, and  is the transmitter beam angular spread.
with an active beam-steering mechanism. Pister and his students This expression assumes that  is small, and that the transmit-
are working on a MEMS-based mechanism capable of steering a ter beam is well-aimed at the receiver. The SNR governs the
beam to any position within a hemisphere. Beam-steering algo- probability of bit error, and must be maintained at a suitably
rithms for systems with active dust mote transmitters represent high value to insure reliable link operation. From (1), we see
that in order to achieve a given SNR with all other parameters
a current research challenge. It would be desirable for each dust
;1=2 , i.e., the
fixed, the required value of E b is proportional to R b
mote to autonomously steer its beam toward the desired direc-
tion. One approach would be to make the dust mote receiver di- energy per bit is minimized if packets are transmitted in short
rectional, and to mount the receiver and transmitter on the same bursts at a high bit rate.
aiming mechanism. Accordingly, by aiming its receiver so as The average transmitter power (during transmission of a
to maximize the signal received from the BTS or another mote, packet) is Pt = Eb =Rb . Hence, transmission at a high bit rate
the dust mote would be aiming its transmitter at that node. The requires a high-power transmitter. In practice, P t should be cho-
need for active dust mote transmitters to determine the direction sen to be as high as possible, within constraints posed by eye
to other nodes slows down connection set up, but if nodes remain safety and by dust mote current-drive limitations. Rewriting (1)
fixed then the directions of various nodes, once determined, can in terms of Pt , we obtain

SNR = C N PRt Ad4 4 :

be stored in the dust mote for future use. 2 2
Under most of the scenarios discussed above, the dust mote’s
(2)
transmitter and receiver have different angular spreads. This 0 b
leads to non-reciprocal link characteristics, wherein a dust mote Given a limit on Pt , to maximize the bit rate R b and the distance
may receive from another node, but be unable to transmit to it, or d, we should maximize the receiver area A and minimize , i.e.,
vice versa. As a consequence, a dust mote may receive queries use a highly directional transmitter.
from other nodes, and may attempt to answer them, unaware that Once all other parameters have been fixed, to maintain a re-
its transmissions are in vain. When dust motes are fixed, in order quired SNR, the permissible bit rate and distance are related by
to conserve dust mote power, the other nodes should acknowl- Rb / d;4 . Hence, it is possible to extend the link distance by
edge this dust mote’s transmissions, and this dust mote should drastically lowering the bit rate. If a multihop route is available,
not answer further queries from nodes that do not acknowledge overall latency may be minimized by transmitting at a higher bit
its transmissions. rate over several hops.
1 In a well-designed CCR, the angular spread of the reflected beam is limited 2 On a link from BTS to dust mote or from dust mote to dust mote, A corre-

by diffraction to the order of  =a, where  is the optical wavelength and a sponds to the dust mote photodetector area. On a link from dust mote to BTS,
is the effective diameter of the CCR. A corresponds to the BTS camera’s entrance aperture area.
 To give a concrete example, the inch-scale laser node shown described for dust motes, the channel is “granted” by the inci-
in Fig. 1 has an average optical power of 1.5 mW and a beam dent probe beam.
divergence of roughly 1 mrad. The receiver used in the 21- Note that there are as many channels (paging or data) as there
km trans-bay demonstration had a 1-inch diameter lens aperture, are resolvable pixels at the BTS. The BTS has no way to dis-
and the signal to noise ratio at 4 bps was more than 30 dB (the tinguish between simultaneously communicating dust motes if
upper bound was not measurable because of pixel saturation and they fall within the same pixel in the imaging array. One possi-
blooming due to excessive signal strength). ble way to deal with this is to introduce time slotted techniques
not unlike that found in time division multiple access (TDMA)
B. Mobile Networking Opportunities communications systems. A wide-aperture beam from the BTS
could be modulated in such a fashion as to offer a common time
B.1 Overview base by which to synchronize the motes. The BTS can then sig-
The optical free-space communication method presents many nal an individual mote the particular time slot it has assigned
opportunities beyond low-power, passive communications. to it for communication. The mote must await its time slot to
Since the application of interest in sensor networks is primar- communicate, whether it uses an active or a passive transmitter.
ily sensor read-out, the key protocol issues are to perform read-
out from a large volume of sensors co-located within a poten- B.4 Probe Revisit Rates
tially small area. Random access to the medium is both energy- Probe beam revisit rates could be determined in an
consuming and bandwidth inefficient. So it is extremely use- application-specific manner. It is a well known observation from
ful to exploit passive and broadcast-oriented techniques when statistical data management that areas where changes are hap-
possible. Fortunately the free-space approach supports multi- pening most rapidly should be revisited most frequently. If sen-
ple simultaneous read-out of sensors, mixes active and passive sor readings are not changing much, then occasional samples are
approaches using demand access techniques, and provides effi- sufficient to obtain statistically significant results. So it is better
cient and low-latency response to areas of a sensor network that to spend probe dwell time on those sensors that are experiencing
are undergoing frequent changes. These are described in more the most rapid reading changes, and for which infrequent visit
detail in the following subsections, with emphasis on passive would lead to the greatest divergence from the current sensor
dust mote transmitters. values.

B.2 Parallel Read-Out

IV. APPLICATONS
A single wide beam from the BTS can simultaneously probe
A. Introduction
many dust motes. The imaging receiver at the BTS receives
multiple reflected beams from the motes, as long as they are Depending on the application, individual dust motes may be
sufficiently separated in space to be resolved by the receiver’s affixed to objects that one wishes to monitor, or a large collec-
pixel array. The probe beam sweeps the three dimensional space tion of motes may simply be dispersed (and floating!) at random
covered by the base station on a regular basis, most likely deter- throughout an environment. The motes record sensor readings
mined by the nature of the application and its need for moment- and, when queried, report these readings via the optical tech-
by-moment sensor readings. niques described in Section II. In some applications, dust motes
will communicate directly (and passively) with the BTS, in oth-
B.3 Demand Access ers, peer-to-peer active communication between dust motes will
be used to relay information to the BTS. Depending on the ap-
To save transmit power, if the mote must use active communi- plication, the base station may be separated from the dust motes
cations, then it is best to use the active transmitter in a high-bit- by distances ranging from tens of meters to kilometers.
rate, short-burst mode. Familiar demand access methods can be For example, the BTS may actually reside in a hand-held unit,
used to combine the low latency advantages of active communi- much like a pair of binoculars. This permits the user to simul-
cations with the low-power advantages of the passive approach. taneously view a scene while displaying measured data overlaid
When the mote needs to transmit information, it actively on top of it. As another example, the BTS may reside in a small
transmits a short-duration burst signal to the BTS. The BTS, de- flying vehicle, which flies over an area to query the Smart Dust.
tecting this signal, then probes in the general geographical area We envision numerous civilian and military applications for
from which the burst was detected. Assuming that the passive Smart Dust. Smart Dust may be deployed over a region to record
transmitter (i.e., CCR) is properly oriented toward the BTS, the data for meteorological, geophysical or planetary research. It
mote can respond by modulating the reflected probe beam with may be employed to perform measurements in environments
the data it needs to transmit. where wired sensors are unusable or lead to measurement er-
Logically, the communications structure described above has rors. Examples include instrumentation of semiconductor pro-
much in common with familiar cellular and satellite networks cessing chambers, rotating machinery, wind tunnels, and ane-
[5]. The paging channel is acquired using contention access choic chambers. In biological research, Smart Dust may be used
techniques. The BTS grants a channel to the node requesting at- to monitor the movement, habits, and environment of insects or
tention. In a cellular network, this is accomplished by assigning other small animals. Considering the military arena, Smart Dust
a frequency, time slot, and/or code to the node. In the scheme may be deployed for stealthy monitoring of a hostile environ-
ment, e.g., for verification of treaty compliance. Here, acoustic, treated in the same way). We can assume some maximum duty
vibration or magnetic field sensors could detect the passage of cycle between probe visits. If sufficient time has passed since
vehicles and other equipment. Smart Dust could be used for the last visit, the mote can assume that it is blocked. Weighted
perimeter surveillance, or to detect the presence of chemical or by the importance of what it has detected, the mote can decide
biological agents on a battlefield. to go active.
The overarching applications challenge, from a processing Building a multihop route in this environment is quite chal-
and communications viewpoint, is how to implement complex lenging. Because of the directionality of the on-board laser, ac-
“ensemble” behavior from a large number of individual, rel- tive transmission in all directions is not feasible, and we cannot
atively simple sensors. This is sometimes called “beehive,” assume that if a next hop node receives our transmission that we
“swarm,” or “emergent” behavior. A critical enabler is the abil- will be able to receive a transmission from it.
ity for the sensors to communicate their readings with each other Determining true reachability between pairs of motes requires
and with the more centralized intelligent processor residing at a full four phase handshake (“Can you see me?” “Yes, I can see
the base station. Proper design of the network is the key. We you. Can you see me?” “Yes” “Good. We can communicate
describe an applications scenario and some of the technology with each other.”). This must be executed in the context of ap-
challenges to implement such a system in this section. propriate timeouts and made robust to dynamic changes in the
positions of the communicating nodes, which may be floating in
B. Scenario: Multi-Sensor Emergent Behavior the air.
Assuming a static arrangement initially, we propose the fol-
It is useful for sensors to operate in ensembles. Rather than
lowing connectivity discovery algorithm for a questing mote.
implementing a broad range of sensors in a single integrated cir-
We assume that the mote has a unique ID and a finite set of direc-
cuit, it is possible to simply deploy a mixture of different sensors
tions in which it can point its laser. Furthermore each direction
has a unique description, e.g., (x; y ), in the mote’s coordinate
in a given geographical area and allow them to self-organize.
Sensors are typically specialized to detect certain signatures.
system. In the first phase of discovery, the mote iterates through
One kind detects motion, another heat, and a third sound. When
every direction, transmitting in each case “I am ID1, and I am
pointing at location (x; y ) Any mote hearing this broadcast be-
one sensor detects its critical event signature, it makes other
nearby sensors aware of its detection. They then orient their
gins its own scan, transmitting “I am ID2 pointing at location
sensing function in a particular, signature-specific way. For ex-
(i; j ); and I have heard mote ID1 who was pointing at loca-
tion (x; y ).” Upon hearing mote2’s message, mote1 now knows
ample, a simple motion-detecting sensor might cue more sophis-
ticated sensors detecting thermal or other radiation properties.
where to point to talk to mote2, and in a subsequent message
The array, acting as an ensemble, not only performs the opera-
can tell mote2 where to point to respond.
tion of detecting an intruder, but demonstrates more intelligent
processing, by distinguishing between one that is a human and Routing tables can be constructed from such pairwise discov-
another that is a small animal (e.g., the former has a body heat ery of connectivity. However, standard routing algorithms, like
signature spread over a larger volume than the latter). RIP, OSPF, and DVRMP, assume bidirectional and symmetric
A more complex sensor cued in this fashion may then increase links. This will not always be the case for Smart Dust. It may
its own scan rate to obtain a higher-resolution signature, or ded- be possible for mote A to communicate with mode B, but not
icate its detection energy budget into a particular narrow band vice versa. Even if the communications is bidirectional, it need
or a specific direction. These operations have implications for not exhibit the same bandwidth or loss characteristics in both
power consumption. Maximizing detection probability and res- directions.
olution while minimizing power consumption is a key optimiza- Therefore, new routing algorithms must be developed to deal
tion challenge. with the general case of links that are unidirectional and/or
asymmetric in their performance. A strong group at INRIA in
France has been leading the IETF Unidirectional Link Routing
C. Technology Approaches for Realizing the Scenario
Working Group discussions on these issues [6], [7].
There are two ways to construct such a cueing system. The Unfortunately, the current efforts are focusing on support-
first is a centralized scheme. The motion sensor communicates ing high-bandwidth unidirectional links where all nodes have at
with the BTS, which in turn communicates with a nearby heat least low-bandwidth bidirectional links (e.g., a high-bandwidth
sensor. If passive communications techniques can be used, this satellite link superimposed on nodes interconnected via slow-
may well be the most power-efficient way to propagate the de- speed telephone links). Even modifying existing algorithms will
tection information. not help much, since the connectivity among floating dust motes
The centralized/passive schemes cannot be used if the line-of- is dynamic with short time scales. The more general case still
sight path is blocked, or if the probe revisit rate is too infrequent remains to be addressed.
to meet detection latency constraints. In these cases, the detect-
ing mote must employ an active transmitter. If the line-of-sight
D. Other Applications Issues
path is blocked, then the mote will need to use ad hoc, multi-
hop techniques to communicate with the BTS or nearby sensor One possible improvement is to make use of emerging MEMS
nodes. technology for on-board inertial navigation circuits [8] to make
Detecting a blocked path between a mote and the BTS is not sensors more aware of near neighbors even as they drift out of
difficult (note that a blocked path and a disabled BTS can be line-of-sight of the BTS. The BTS can determine the relative
location of dust motes within its field of view. It could then dis- of ensembles of simple nodes. Dealing with partial disconnec-
seminate this “near neighbor information” to motes able to ob- tions while establishing communications via dynamic routing
serve its probe beam. The on-board inertial navigation capabil- over rapidly changing unidirectional links poses critical research
ity, combined with these periodic relative location “snapshots” challenges for the mobile networking community.
could assist motes in orienting their laser and detector optics to
improve their ability to establish links with nearby motes.
ACKNOWLEDGMENTS
V. RELATED PROJECTS
Kahn and Pister’s research is supported in part by DARPA
Several projects have recently been initiated to investigate a Contract DABT63-98-1-0018, “Smart Dust.” Katz and Pister’s
variety of communications research aspects of distributed sensor research is supported in part by DARPA Contract N6601-99-2-
networks. The following description is by no means exhaustive. 8913, “Endeavour.”
The Factoid Project [9] at the Compaq Palo Alto Western Re-
search Laboratory (WRL) is developing a portable device small
enough to be attached to a key chain. The device collects an- REFERENCES
nouncements from broadcasting devices in the environment, and
[1] N. Stephenson, The Diamond Age, Bantam Books, New York, 1995.
these can be uploaded to a user’s home basestation. In its first
[2] V. S. Hsu, J. M. Kahn, and K. S. J. Pister, “Wireless communications
generation, the prototype devices are much larger than smart for smart dust,” Electronics Research Laboratory Memorandum Number
dust motes, communications is accomplished via RF transmis- M98/2, 1998.
sion, and the networking depends on short-range point-to-point [3] P. B. Chu et al., “Optical communication using micro corner cube reflec-
tors,” in Proc. EEE MEMS Workshop, Nagoya, Japan, Jan. 1997, pp. 350–
links. 355.
The Wireless Integrated Network Sensors (WINS) Project [4] F. Gfeller and W. Hirt, “Robust wireless infrared system with channel reci-
[10] at UCLA is very similar in spirit to what has been described procity,” IEEE Commun. Mag., vol. 36, no. 12, pp. 100–106, Dec. 1998.
in this paper. It is developing low power MEMS-based devices [5] D. Goodman, Wireless Personal Communication Systems, Addison-
Wesley Longman, Reading, MA, 1997.
that in addition to sensing and actuating can also communicate. [6] W. Dabbous, E. Duros, and T. Ernst, “Dynamic routing in networks with
The essential difference is that WINS has chosen to concentrate unidirectional links,” Workshop on Satellite-Based Information Systems,
on RF communications over short distances. Budapest, Sept. 1997.
[7] Unidirectional Link Routing Protocol Working Group Home Page, avail-
The Ultralow Power Wireless Sensor Project [11] at MIT is able at http://www-sop.inria.fr/rodeo/udlr/.
another project that focuses on low power sensing devices that [8] B. Boser, “Electronics for micromachined inertial sensors,” Transduc-
also communicate. The primary thrust is extremely low power ers’97, Chicago, Il., pp. 1169–1172, June 1997.
operation. The prototype system will transmit over a range of [9] Available at http://www.research.digital.com/wrl/projects/Factoid/index.
html.
data rates, from 1 bps to 1 Mbps, with transmission power lev-
els that span from 10 W to 10 mW. The RF communications
[10] Available at http://www.janet.ucla.edu/WINS.
[11] Available at http://www-mtl.mit.edu/j̃img/project top.html.
subsystem is being developed for the project by Analog De- [12] J. Jubin and J. D. Turnow, “The DARPA packet radio network protocols,”
vices. Again, optical technologies are not being investigated. in Proc. IEEE, vol. 75, no. 1, Jan. 1987, pp. 21–32.
Ultimately the design team will need to face the multi-hop wire- [13] G. S. Lauer, Packet-radio networks, ch. 11 in Routingin Communications
Networks, M. Steenstrup, Ed., Englewood Cliffs, Prentice-Hall, N.J.,1995.
less networking protocol issues outlined in this paper (e.g., see
[12], [13]).

VI. SUMMARY AND CONCLUSIONS

The research community is searching for a new environments
in which to generate innovative ideas and prove their effective- Joseph M. Kahn is a Professor in the Department of
ness. A new paradigm beyond desktop computing is capturing Electrical Engineering and Computer Sciences at U.C.
the imaginations of systems designs: the so-called “post-PC” Berkeley. In 2000, he co-founded StrataLight Com-
munications, Inc., where he is currently Chief Scien-
era. Wireless sensor networks is one area that promises to yield tist. He received the A.B., M.A., and Ph.D. degrees
important applications and demands new approaches to tradi- in Physics from U.C. Berkeley in 1981, 1983, and
tional networking problems. 1986, respectively. From 1987–90 he was a Mem-
ber of Technical Staff in the Lightwave Communica-
We have described Smart Dust, an integrated approach to net- tions Research Department of AT&T Bell Laborato-
works of millimeter-scale sensing/communicating nodes. Smart ries, where he performed research on multi-gigabit-
per-second coherent optical fiber transmission sys-
Dust can transmit passively using novel optical reflector tech- tems, setting several world records for receiver sensitivity. He joined the fac-
nology. This provides an inexpensive way to probe a sensor ulty of U.C. Berkeley in 1990. His current research interests include wireless
or acknowledge that information was received. Active optical communication using antenna arrays, free-space optical communication, opti-
cal fiber communication, and wireless communication for networks of sensors
transmission is also possible, but consumes more power. It will based on micro-electromechanical systems. Professor Kahn received the Na-
be used when passive techniques cannot be used, such as when tional Science Foundation Presidential Young Investigator Award in 1991. He
the line-of-sight path between the dust mote and BTS is blocked. is a Fellow of the IEEE, and is a member of the IEEE Communications Society,
the IEEE Information Theory Society, and the IEEE Lasers and Electro-Optics
Smart dust provides a very challenging platform in which to Society. From 1993–2000, he served as a technical editor of IEEE Personal
investigate applications that can harness the emergent behavior Communications Magazine.
 Randy Howard Katz received his undergraduate de- Kristofer S. J. Pister received his B.A. in Applied
gree from Cornell University, and his M.S. and Ph.D. Physics from UCSD in 1982, and his M.S. and
degrees from the University of California, Berkeley. Ph.D. in Electrical Engineering from UC Berkeley in
He joined the faculty at Berkeley in 1983, where he is 1989 and 1992. From 1992 to 1997 he was an As-
now the United Microelectronics Corporation Distin- sistant Professor of Electrical Engineering at UCLA,
guished Professor in Electrical Engineering and Com- where he developed three graduate level courses in
puter Science. He is a Fellow of the ACM and the Micro Electro Mechanical Systems: MEMS device
IEEE, and a member of the National Academy of En- physics and fabrication, MEMS system design, and
gineering. He has published over 180 refereed tech- CAD for MEMS. In 1996 he joined the faculty of
nical papers, book chapters, and books. His hard- Electrical Engineering and Computer Sciences at UC
ware design textbook, Contemporary Logic Design, Berkeley as an Associate Professor. From 1996 to
has sold over 85,000 copies worldwide. He has supervised 32 M.S. theses and 1999 he was a member of the DARPA ISAT group, and in 1997 he was chosen
17 Ph.D. dissertations. He has won numerous awards, including seven best pa- to be a part of the Defense Science Study Group. During the last five years, his
per awards, one “test of time” paper award, three best presentation awards, the primary research interest has been the development and use of standard MEMS
Berkeley Distinguished Teaching Award, the 1999 IEEE Reynolds Johnson In- fabrication technologies, micro robotics, and CAD for MEMS. He is the in-
formation Storage Award, the 1999 ASEE Frederic E. Terman Award, and the ventor of the polysilicon hinge, now in use by many MEMS groups around the
1999 ACM Karl V. Karlstrom Outstanding Educator Award. With colleagues at world. Most recently his work has been in wireless sensor networks and micro
Berkeley, he developed Redundant Arrays of Inexpensive Disks (RAID), an $18 optics. He is an active consultant in the MEMS industry.
billion per year industry sector today. While on leave for government service
in 1993–1994, he established whitehouse.gov and connecting the White House
to the Internet. His current research interests are Internet Services Architecture,
Mobile Computing, and Computer-Telephony Integration.