Exploring System Energy Limits with Smart Dust

Brett Warneke, Kristofer Pister
Berkeley Sensor and Actuator Center
University of California, Berkeley
{warneke,pister}@eecs.berkeley.edu

The Problem

    The DARPA-funded Smart Dust project aims to explore the limits of miniaturization by packing an autonomous sensing, computing, and communication system into a cubic millimeter mote that will form the basis of massive distributed sensor networks, thus demonstrating that a complete useful, yet complex, system can be integrated into 1mm3.  Because of the discreet size, substantial functionality, connectivity, and expected low cost, Smart Dust will enable entirely new methods of interacting with the environment, providing more information from more places in a less intrusive way than ever before.  Some examples of things that we are pursuing include defense networks that could be rapidly deployed by unmanned aerial vehicles (UAV) or artillery,  tracking the movements of birds, small animals, and even insects, virtual keyboards, inventory control, product quality monitoring, smart office spaces, and interfaces for the disabled.

    Smart Dust will require both evolutionary and revolutionary advances in miniaturization, integration, and energy management.  These advances will be facilitated by the progress in micro electromechanical systems (MEMS), which allows us to build small sensors, communication components, and power supplies, and microelectronics, which provides increasing amounts of functionality in smaller areas and with lower energy consumption.  The power system may consist of a thick film battery and/or a solar cell with a charge integrating capacitor for periods of darkness.  A variety of sensors, including light, temperature, vibration, magnetic field, acoustic, and wind shear, can be integrated on the mote depending upon the mission.  An integrated circuit will provide sensor signal processing, communication, control, data storage, and energy management.

    Because of the small size of the mote, energy management is a key component of the design.  Current battery and capacitor technology can store approximately 1J/mm3 and 10mJ/mm3, respectively, while solar cells can provide 1J/day/mm2 in sunlight and 1-100mJ/day/mm2 indoors.  Energy consumption must therefore be minimized in every part of the system.
 

The Solution

    In order to meet our aggressive size and energy constraints, free space optical communications will be utilized because it allows large antenna gain and small radiators.  The large antenna gain, a lack of multi-path fading, and the lower overhead of laser diodes as compared to RF power amplifiers provide significantly less energy consumption per bit to transmit a given distance.  Furthermore, laser diodes and micromachined mirrors can be much smaller than RF antennas.  Utilizing a CMOS photodiode, we are designing an optical receiver front-end that should consume approximately 0.1nJ/bit.  For data transmission, two schemes are being explored: passive transmission using a corner-cube retroreflector that will only consume on the order of 1nJ/bit and active transmission using a laser diode and steerable mirrors.

    The mote controller will provide back-end communications processing, sensor sampling control and signal processing, data storage, and energy management.  In addition, because of the difficulties imposed by the size of the mote the processor needs to be reprogrammable over the communications channel.  We are therefore designing new low energy processors that meet the needs of the mote.  One architecture is a somewhat traditional controller/datapath organization, but a new architecture is also being designed that provides substantial energy savings through a better matching between the architecture and the application space.  Since the mote is very dependent on the environment, yet even checking the environment for signals is expensive energy-wise, we utilize a reactive architecture that employs reconfigurable datapath components that are set up to execute an operation only when required by timers that control the frequency of the polling.
 

Justification

    Our miniaturization techniques have been demonstrated with a 138mm3 autonomous uni-directional sensing/communication mote that optically transmits a measure of the incident light level.  Furthermore, we have developed a 63mm3 autonomous bi-directional communication mote that receives an optical signal, generates a pseudorandom sequence based on this signal to emulate sensor data, then optically transmits the result.  These motes are dominated by the power source, which are the smallest button cells available on the market.  Based on their power consumption of 17µW, if we switch to solar cells we will be near our target volume of a cubic mm.

    Our new controller architecture shows promise of being substantially superior for our application than competitive microcontroller architectures.  One of the most efficient microcontrollers available is the CoolRisc 81 core that uses 22pJ/instruction, as compared to the StrongARM SA1100 that consumes ~1nJ/instruction.  To perform a typical sequence of operations, preliminary simulations predict that our new architecture will use 4.8pJ, while the CoolRisc 81 would require 136pJ.

    These demonstrations indicate that we are successfully exploring the limits of complex system miniaturization and energy efficiency, while producing a device that will change the way we think about gathering data from the environment.