ABSTRACT
The Smart Dust project aims to explore the limits of system integration by packing an autonomous sensing, computing, and communication node into a cubic millimeter mote that will form the basis of massive distributed sensor networks, thus demonstrating that a complete system can be integrated into 1 mm3. Effectively exploring this space requires new approaches to design that emphasize energy and volume constraints over all others. To this end a 16 mm3 autonomous solar-powered sensor node with bi-directional optical communication has been demonstrated, with smaller nodes forthcoming. System integration limits will shrink even further as carbon nanotube technology matures.
INTRODUCTION
In recent decades there have been dramatic reductions in the size of computational devices, sensors, and wireless communication. In addition, there has been an increasing amount of integration among these three areas and power supplies, creating systems with more functionality in smaller packages. Extrapolating these trends, we envision wireless sensor nodes becoming as small and as numerous as dust --disappearing into the environment and dramatically changing the manner in which we interact with it. The Smart Dust project [1] was begun with the goal of investigating the limits of miniaturization and system integration by demonstrating that a useful and complex system can be built within a cubic millimeter. Specifically, the project is seeking to build an autonomous device that incorporates sensors, computation, wireless communication, and power source in a cubic millimeter mote that can be used in a distributed sensor network. Such
devices would provide more information from more places in a less intrusive manner than ever before. Some example applications include defense networks that could be rapidly deployed by unmanned aerial vehicles (UAV), tracking the movements of birds, small animals, and even insects, fingertip accelerometer virtual keyboards, monitoring environmental conditions affecting crops and livestock, inventory control, and smart office spaces.
The development of Smart Dust has required advances in miniaturization, integration, and energy management. This paper will discuss the philosophy of design that has been necessary to really explore this space, describe the mote architecture and components, show the current realizations of the system, then briefly discuss future limits.
DESIGN PHILOSOPHY
To really push the limits one must first eliminate any commitment to existing standards, protocols, packaging techniques, etc. that can add overhead and decrease efficiency. While these standards have their place and have played an important role in the success of certain technologies, they can prevent one from really approaching the limits and thus opening up new applications. For example, the IEEE 802.11b wireless network standard was designed for the high data rates needed by a computer network, but with some power consumption considerations since it was intended to be used in battery-powered devices such as laptop computers. Wireless sensor networks, on the other hand, have a substantially different usage model that includes much lower data rates and more spurious traffic that can lead to a dramatic reduction in the energy required to communicate a certain amount of information.
Exploring the Limits of System Integration with Smart Dust
Brett A. Warneke, Kristofer S.J. Pister
University of California, Berkeley Berkeley Sensor and Actuator Center
497 Cory Hall Berkeley, CA 94720-1774 {warneke,pister}@eecs.berkeley.edu
Proceedings of IMECE’02 2002 ASME International Mechanical Engineering Congress & Exposition New Orleans, Louisiana, November 17-22, 2002
Nevertheless, larger bridge nodes will exist to take advantage of network and data standards by interfacing with a wireless sensor network.
A second design perspective that is critical when miniaturizing systems is that operations should be thought of in terms of energy. Because of the diminutive size of these devices and their wide dispersion, it will not be practical to change or recharge any batteries. Therefore, the energy stored in the battery limits the life of the mote and we can break down each operation into a particular energy cost to better guide our design decisions. Table 1 delineates this concept by showing the energy cost of various operations and how many operations a cubic millimeter of battery would yield. For example, an eight bit microprocessor instruction on a Smart Dust mote consumes less than 10pJ, so a 1J battery would only provide a little over 100 billion operations over its lifetime. Furthermore, one can translate component volumes into energy opportunity costs. With these figures, one can make trade-offs between computation, sensor sampling, communication, and volume. For instance, it may be advantageous to spend extra energy compressing data in order to save more energy during communication. One may also want to adjust the size of the transmitter and thus affect its energy efficiency as a trade-off with the size of the battery.
THE SMART DUST ARCHITECTURE
Figure 1 shows a conceptual diagram of a Smart Dust mote. As discussed above, the volume is expected to be dominated by the energy source, which will be an energy storage element such as a battery or capacitor, an energy harvester, or a combination thereof. A number of approaches to creating microbatteries are currently being pursued, including thick films, micromachined thin films [2], and micromachined cavities with an electrolyte [3]. From the system’s perspective, a good microbattery would have the following features:
1. high energy density
2. large active volume to packaging volume ratio (i.e. a thin film on top of a 500µm silicon wafer would not be desir-able)
3. small cell potential (0.5 - 1.0 V) so digital circuits can take
9 years (1 set/sec) 10 years 20 years 26 years 32 years Battery lifetime @ 100 ops/sec 63 million 63 billion 16 pJ nJ TX 1 bit 83 million 83 billion 12 pJ nJ RX 1 bit 32 million 32 billion 31pJ nJ Sensor sample 300,000(every 0.3 sec) 300 million 3 nJ 200 nJ Sample, think, listen, talk >100 million >100 billion <10 pJ nJ Digital inst. Ops/day*mm2 indoor solar Ops/mm3 battery Now Previous Operation 9 years (1 set/sec) 10 years 20 years 26 years 32 years Battery lifetime @ 100 ops/sec 63 million 63 billion 16 pJ nJ TX 1 bit 83 million 83 billion 12 pJ nJ RX 1 bit 32 million 32 billion 31pJ nJ Sensor sample 300,000(every 0.3 sec) 300 million 3 nJ 200 nJ Sample, think, listen, talk >100 million >100 billion <10 pJ nJ Digital inst. Ops/day*mm2 indoor solar Ops/mm3 battery Now Previous Operation
Table 1: Energy Consumption and Battery Life advantage of the quadratic reduction in power consumption with supply voltage
4. efficiently configured into series batteries to provide a vari-ety of cell potentials for the various components of the sys-tem without requiring the overhead of voltage converters 5. rechargeable in case the system has an energy harvester Microbattery technology is still heavily in the research phase, but current micromachined batteries can provide 5.6 J/mm3[2], which is competitive with large-scale batteries. Even though process compatibility with the other components of the system may seem desirable, it may actually not be important because of the possibility of stacking the components.
Scavenging energy from the environment will allow the wireless sensor nodes to operate nearly indefinitely, without their battery dying. Solar radiation is the most abundant energy source and yields around 1 mW/mm2 (1 J/day/mm2) in full sunlight or 1 µW/mm2under bright indoor illumination. Solar cells have conversion efficiencies up to 30% and are a well-established technology, making them attractive for early use in sensor nodes. Vibration harvesting [4] is another potential energy source, scavenging energy from the vibrations of copy machines, ventilation systems, etc. More exotic energy sources include utilizing the excess heat from micro rocket engine combustion [5] and micro radioactive sources.
Capacitors may be used in these systems to effectively lower the impedance of a battery or energy harvester to allow larger peak currents or to integrate charge from a energy harvester to compensate for lulls, such as nighttime for a solar cell. Current capacitors store up to 10 mJ/mm3.
The next component of the Smart Dust mote is the sensor array. Micromachining has allowed researchers to shrink many types of sensors into small volumes while often maintaining similar, or even exceeding, performance levels of conventional transducers [6,7]. Monolithically integrating sensors together or
Figure 1. Smart Dust conceptual diagram.
1-2mm
Thick-Film Battery Solar Cell
Power Capacitor Analog I/O, DSP, Control
Sensors
Passive Transmitter with Corner-Cube Retroreflector
Interrogating Laser Beam
Mirrors
Active Transmitter with Beam Steering
Laser Lens Mirror
Photodetector and Receiver
Incoming Laser Communication 1-2mm Thick-Film Battery Solar Cell Power Capacitor Analog I/O, DSP, Control
Sensors Sensors
Passive Transmitter with Corner-Cube Retroreflector
Interrogating Laser Beam
Mirrors
Passive Transmitter with Corner-Cube Retroreflector Interrogating Laser Beam Mirrors Interrogating Laser Beam Mirrors Active Transmitter with Beam Steering
Laser Lens Mirror
Active Transmitter with Beam Steering
Laser Lens Mirror Laser Lens Mirror
Photodetector and Receiver
Incoming Laser Communication
Photodetector and Receiver
Incoming Laser Communication Incoming Laser Communication
with other components such as the circuits can be advantageous because of the reduction in parasitics and area lost to pads, but placing them on disparate chips can be advantageous for stacking and packaging concerns of the micromachined structures.
Free-space optical communication provides several advantages over RF communication for small, energy-constrained wireless nodes. First, optical radiators can be made more efficient as well as with much higher antenna gain (> 106) at the millimeter scale. Furthermore, optical transmitters are more power efficient at low power because of reduced overhead and since received power only drops as1/d2, compared with1/ d4for RF transmissions subject to multi-path fading.
Two compact approaches to free-space optical transmission include passive reflective systems and active-steered laser systems. The passive reflective system consists of three mutually orthogonal mirrors that form the corner of a cube (Fig. 1), hence the name corner cube retroreflector (CCR). Light entering the CCR bounces off each of the mirrors and is reflected back parallel to the incident beam. By electrostatically actuating the bottom mirror, the orthogonality can be broken, causing less light to return to the sender. The CCR can thus communicate with an interrogator by modulating the reflected light, with the only energy consumption being the charging of about 3pF in the actuator and a demonstrated range of 180m [8]. This technique consumes much less power than an approach that requires the generation of radiation, such as lasers or RF, but it does not facilitate peer-to-peer communication.
Active-steered laser communication is currently under development. It would utilize an onboard light source, such as a VCSEL, a collimating lens, and MEMS beam-steering optics [9,10] to send a tightly collimated light beam toward an intended receiver, thus facilitating peer-to-peer communication. The final major aspect of the mote are the integrated circuits that tie all the pieces together. Analog signal conditioning circuitry and an analog to digital converter provide the interface to the sensors, while the receiver utilizes an integrated photodiode to sense optical transmission then decodes the data and timing information. Finally, a microcontroller orchestrates all of the functions of the mote, manages energy consumption, and stores data in an SRAM. All of these circuits are designed from the beginning to consume the minimum amount of energy to perform their given task.
SYSTEM INTEGRATION
At the top of the system stack is the software that controls the motes. TinyOS, an operating system developed for wireless sensor networks, is event-driven and supports efficient modularity and concurrency-intensive operation. The modularity is particularly important from a system integration perspective because the software modules in the operating system can be easily swapped for hardware components, and
vice versa, allowing flexibility in the system components. This software is currently running on platforms such as is shown in Fig. 2, which is a cubic inch-scale mote built with commercial off-the-shelf components. TinyOS will eventually be ported to the cubic millimeter motes
The current generation of Smart Dust mote (Fig. 3) is a 16mm3 autonomous solar-powered sensor node with bi-directional optical communication. The device digitizes integrated sensor signals and transmits and receives data optically. The system consists of three die–a 0.25µm CMOS ASIC, a trench-isolation SOI solar cell array, and a micromachined four-quadrant CCR. The dramatic size difference between the devices in Fig. 2 and Fig. 3 show the benefits of designing for low energy consumption and small size from the beginning. The next generation device will incorporate an ultra-low energy custom microcontroller and SRAM into the CMOS ASIC. Currently the mote is hand assembled, but microassembly techniques including microdomain pick and place and flip-chip bonding could be used to automate the process and even make the system more compact. Furthermore, process integration will combine the CCR, solar cell array, and accelerometer die into one in the next generation mote, shrinking it down to 6.6mm3and simplifying the assembly process.
Packaging a cubic millimeter device obviously requires some innovative solutions. Some of the requirements are:
1. Protect the microstructures such as the CCR, accelerometer, and bond wires, while still allowing them to move. 2. Solar cells - clear packaging and possibly a lens 3. Receiver photodiode - optical filter
Figure 2. TheMicamote built from COTS
components. It incorporates a coin cell battery, a radio transceiver, microcontroller, various sensors, and a module connector that allows a wide
4. CCR - anti-reflective (AR)-coated cover that allows illumi-nation along its primary axis of [111].
5. Not add much extra volume
The best proposed solution at this time incorporates potting the mote in an optical-quality polymer with some special molds.
EXPLORING THE LIMITS IN THE FUTURE
While Smart Dust is approaching a cubic millimeter mote with today’s micro technology, future nanotechnology could conceivably push the limits of system integration size down
Figure 3. Mock-up of the 16mm3autonomous
solar-powered mote with bi-directional communications and sensing, composed of a 0.25µm CMOS ASIC, solar power array, accelerometer (not yet demonstrated in the system) and CCR, each on a separate die.
Solar Cell Array
CCR
XL
CMOS
IC
Photosensor
Receiver
even further. Of the four components necessary in a mote (sensing, computation, communication, and power), both sensing and computation have been demonstrated with carbon nanotubes [12,13]. Nanotubes and other nanotechnology devices may also play an important role in radio communication (as oscillators, filters, mixers, etc.), and power storage (hydrogen storage) [14] and generation (e.g. nano fuel cells).
CONCLUSIONS
Smart Dust has shown that system integration can be pushed down into the cubic millimeter-scale while maintaining a complex and useful system that incorporates computation, sensing, communication, and energy source. Further developments in microtechnology can reduce the size by perhaps a factor of ten, but nanotechnology should provide the next leap in the reduction of the limits.
REFERENCES
[1] B. Warneke, M. Last, B. Leibowitz, and K.S.J Pister, K.S.J., 2001, “Smart Dust: Communicating with a Cubic-Millimeter Computer”,Computer Magazine, Jan. 2001, pp. 44-51.
[2] W.C. West,et al., “Fabrication and testing of all solid-state microscale lithium batteries for microspacecraft applications,”
J. Micromechanics and Microengineering, 12(2002), p. 58-62.
[3] K.B. Lee, L. Lin, “Electrolyte Based On-Demand and Disposable Microbattery,”MEMS 2002, Las Vegas, Nevada, 20-24 Jan. 2002, p.236-239.
[4] S. Meninger,et al., “Vibration-to-electric energy conversion,”IEEE Trans. VLSI Systems, vol.9, (no.1), Feb. 2001, p.64-76.
[5] D. Teasdale, V. Milanovic, P. Chang, K.S.J. Pister,
“Microrockets for Smart Dust,”Smart Materials and Structures, vol.10, (no.6), Dec. 2001, pp.1145-55.
[6] Lj. Ristic [ed],Sensor Technology and Devices, Artech House, London, 1994.
[7] G.T.A. Kovacs,Micromachined Transduceers Sourcebook, WCB McGraw-Hill, San Francisco, 1998.
[8] L. Zhou, K.S.J. Pister, J.M. Kahn, "Assembled Corner-cube Retroreflector Quadruplet",MEMS 2002, Las Vegas, Nevada, 20-24 Jan. 2002, p.556-559.
[9] M. Last,et al., “An 8 mm3digitally steered laser beam transmitter,”2000 IEEE/LEOS Int’l Conf. Optical MEMS, Kauai, HI, p. 69-70.
[10] V. Milanovic, M. Last, K.S.J. Pister, ``Torsional Micromirrors with Lateral Actuators,''Trasducers'01
-Eurosensors XV Conference, Muenchen, Germany, Jun. 2001.
[11] J. Hill, R. Szewczyk, A. Woo, S. Hollar, D. Culler, K. Pister, “System architecture directions for networked sensors,”
ASPLOS-IX. Ninth International Conference on Architectural Support for Programming Languages and Operating Systems, Cambridge, MA, USA, 12-15 Nov. 2000, p.93-104.
[12] M.S. Fuhrer, J. Nygard, L. Shih, M. Forero, Young-Gui Yoon, M.S.C. Mazzoni, Hyoung Joon Choi, Jisoon Ihm, S.G. Louie, A. Zettl, P.L. McEuen, “Crossed nanotube junctions.”
Science, vol.288, (no.5465), 21 April 2000. p.494-7.
[13] Jing Kong, N.R. Franklin, Chongwu Zhou M.G. Chapline, Shu Peng, Kyeongjae Cho, Hongjie Dai, “Nanotube molecular wires as chemical sensors,”Science, vol.287, (no.5453), 28 Jan. 2000. p.622-5.
[14] Wang Qikun, Zhu Changchun, Liu Weihua, Wu Ting, “Hydrogen storage by carbon nanotube and their films under ambient pressure,”International Journal of Hydrogen Energy, vol.27, (no.5), May 2002. p.497-500.
