Technology was on duty miles downwind, where, as the battle against these fires went on for days, even weeks, many residents of the Bay Area and Sacramento River Delta region turned to air-quality sensor networks, particularly
AirNow, maintained by the U.S. government, and PurpleAir, created via crowd-sourcing of commercial sensors. The data from these two broad sensor networks helped residents decide whether to wear an N95 particle mask when going outside, whether it was safe to exercise or to let children play outdoors, how long to keep the air filters running inside the house, and how far to drive to escape.

These particular networks use
sensor units mounted on buildings to stream data via Wi-Fi to Web-based mapping programs. Just a few hundred sensor units distributed over the larger Bay Area were enough to identify significant local differences in the spread of smoke. For instance, the sensors showed that the topography of the Santa Cruz Mountains protected downwind coastal towns from smoke, while the Sacramento River Delta suffered far more as smoke stagnated in its wide, low areas.

It’s great that sensors tracked smoke in these areas. But why weren’t they on the job where they were really needed, where these wildfires started, to issue an alert before the fires spread?

The main reason is access to power. Sensors that mount on buildings can just plug into a wall outlet. A sensor system that could detect a fire started in a forest does not have that luxury.

Could it use batteries instead, at least one per sensor node?

A woman is running in the foreground. Behind her is a tan house. In the background, flames and smoke can be seen approaching.
A resident of Vacaville, Calif., was one of many Northern Californians forced to flee the LNU Lightning Complex fires in August 2020, after an unusual series of thunderstorms sparked nearly 400 blazes.Philip Pacheco/Bloomberg/Getty Images

Pause for a moment to look at the smoke detector in the room where you’re sitting and think about the last time you changed its battery. Kind of a pain, wasn’t it? A sensor network that could monitor an entire forest, or a gas pipeline, or any critical infrastructure, would need thousands or even millions of sensors—and batteries. Just thinking about the crew of people needed to tramp around to change all those batteries is exhausting, and to actually do it would be prohibitively expensive and impractical.

If we had a sensor network that rarely—or never—drew power, imagine how many important places and things could be monitored, how many lives could be saved. Consider bridges and dams that could report on their structural integrity. Or think about city streets that could report storm flooding, or downed power lines that could identify the exact location of the break and possible risk of fire.

Before we talk about how we might create such a zero-power monitoring system, let’s review the basic components of a distributed sensor network. Besides the power source and the sensors themselves, each node in the network requires a computer (in the form of a microprocessor or a microcontroller chip) and a radio. Typically, the computer is in control: It accumulates sensor data at specific intervals and processes the data. Then it turns on the radio to transmit the data. If the power source is limited in capacity, such as a battery, or in availability, such as a solar panel, the computer also monitors and manages power consumption.

When we talk about managing power consumption here, we usually focus on the power used by the radio. A radio can be very power hungry; the farther a radio signal needs to reach, the more power it must draw.

For those PurpleAir and the other building-mounted sensors mentioned, the radio signal needs to reach just several meters, to a base station, potentially using a low-energy radio protocol like Bluetooth Low Energy or Zigbee, or to an Internet router using Wi-Fi. Out in the forest, though, that’s not the case. Even with mesh networking—a protocol that allows messages to be passed in short hops from node to node on the way back to home base—a large-area network might require each node to transmit over kilometers. To reach such long distances, each radio could need watts, versus only the milliwatts of power available in Bluetooth Low Energy.

One way to conserve power is by programming the computer to sample and transmit on fixed time intervals, say once per hour. Or it might continuously monitor the sensor’s output data and transmit data only when something interesting happens, such as when a prescribed sensor threshold level has been exceeded. But in either case the computer must always be running, which means it will eventually drain the battery.

The ideal sensor warning system, like that pet dog guarding a home at night, would normally remain asleep; however, a certain threshold of noise or smell will cause it to wake up and start barking a warning.

A much better way to conserve battery power would be to use none of it at all until the system actually had important data to transmit. The system would remain in an ultralow-power sleep mode, or even an open-circuit mode, with no current flowing, until the sensor itself detected an important signal.

In this vision, the sensor is in control, not the computer. The sensor would trigger the computer to power up, process the data, and transmit it. And then, with transmission complete and the triggering stimulus gone, the system would shut down and return to a sleep or fully powered-off state. Sleep mode, or something close to it, already appears in virtually every modern IC—particularly those intended for use in mobile devices, where conserving battery life is critical.

The ideal sensor warning system, like that pet dog guarding a home at night, would normally remain asleep; however, a certain threshold of noise or smell will cause it to wake up and start barking a warning.

The sensor equivalent of a sleeping dog is called an event-driven sensor. In its most common form, it uses an incoming stimulus, at some minimum threshold value, to move and close a mechanical switch, which in turn activates an electronic circuit. Once the switch closes, the circuit draws power from the battery and then performs more power-intensive duties like data processing and radio transmission.

Using microelectromechanical systems (MEMS) technology, we can make such event-driven sensors on silicon chips that are only millimeters in size. Tiny forces can actuate them and thus power electronic circuits embedded within the silicon.

Northeastern University, in Boston, Matteo Rinaldi’s group has demonstrated an event-driven sensor that could help detect a forest fire by reacting to the infrared light emitted from a hot object. On its surface, the sensor has an array of nanoscale metal squares that selectively absorb light from specific wavelengths, causing the sensor to heat up. At a predetermined temperature threshold, the absorbed heat will deform a metal finger that mechanically closes an electrical switch. The mechanism is similar to that used in older home thermostats, albeit at a much smaller scale. Once the stimulus is removed, the metal finger reverts to its original shape and the switch opens.

A gold shiny circle with metallic probes and equipment. A white-gloved hand holds tweezers that hold a small black-and-green rectangular sensor.
This sensor from Northeastern University researcher Matteo Rinaldi sleeps in an ultralow-power mode until infrared light, like that from a fire or hot object, wakes it up. A warning system using this type of sensor could go a decade without a battery change.Matthew Modoono/Northeastern University

By changing the geometry of the absorber and the mechanical switch, you could customize this sensor to respond to different wavelengths and light intensities. It could therefore be used in a sensor network to watch for the heat signature created by a forest fire, or in a security application to look for the hot exhaust from a certain vehicle type passing by. During its inactive state, it draws nearly zero power, having a leakage current of only nanoamperes. This sensor could last for years on its original battery while waiting for a triggering event.

At the
University of Texas at Dallas, Siavash Pourkamali’s group has taken a different approach. They developed an event-driven DC accelerometer that can detect change in tilt. This could be used as a security device, to set off an alarm if an object is moved, or as a package shipping monitor, to determine if a package is upended during transport. Deployed in a sensor network, it could also detect small angle changes in large structures, such as fences, pipelines, roadways, or bridges, indicating potentially troublesome deformation or cracking.

The idea behind this motion event-triggered sensor isn’t new. A hundred years ago, centimeter-scale tilt switches used a conductive blob of mercury rolling along a glass tube to close an electric circuit. The MEMS version, of course, is only a few millimeters in size, and instead of mercury, it uses a suspended block of silicon. When the angle changes, the displaced block closes an electrical circuit. This sensor can be customized to designated tilt thresholds, and it consumes no power while waiting for the triggering motion.

Both of these event-driven sensors still require a battery to power up the rest of the system after a triggering event occurs. The awakened computer must then process the sensor data and begin radio transmission according to its programmed instructions.

With parsimonious use, the battery could last for years, but at some point it will run out. The ultimate dream, therefore, would be to have no batteries at all.

As impossible as that may sound, battery-free sensors already exist. We can create them by using a commonplace technology: radio frequency identification. An RFID tag can be a passive electronic device, with no power source of its own. Instead, it draws power inductively from an external device, called a reader. The reader emits electromagnetic energy across a distance, which couples to the RFID tag’s antenna and generates a transient electric current within the RFID tag’s circuit. This temporary coupling of the reader and tag enables small bits of information to be transmitted, such as a serial number or an account balance. A typical use of RFID in this manner is electronic toll collection; the passive RFID tag resides on the car’s windshield, and the car drives under a reader mounted to an overhead gantry.

Getting to zero-power sensors is well worth the effort and expense; deploying them to warn of wildfires would alone justify the R&D investment.

RFID technology can be used to return a sensor reading, instead of just a tag number. Indeed, it has already been used for years in implanted medical sensors, such as the
CardioMEMS system. In that system, a glass-based MEMS capacitive pressure sensor within an aortic aneurysm stent allows a cardiologist to check for stent leakage by placing a reader against the patient’s torso.

But there’s a lot more that can be done with RFID-style powering and readout.

Tsinghua University, in Beijing, Zheng You’s group developed an acoustic-wave sensor that can passively detect temperature change with precision. This device relies on the fact that the center frequency of a piezoelectric structure shifts with variations in temperature, and small frequency shifts can be easily detected by the RFID reader’s circuitry.

With the addition of a chemically selective absorbing coating to the piezoelectric surface, the sensor could measure the concentration of a gas. As the coating absorbs the target gas molecules, the mass resting on the piezoelectric material would increase, again shifting the resonant frequency.

Any sensor that can convert a physical phenomenon into a change in resonant frequency could be read by RFID and therefore operated without a battery. In this case, the challenge involves getting the reader close enough to each and every sensor in the network. It’s hard to imagine doing this for a forest-fire detection system. Putting a larger antenna on the sensor, as well as on the reader, would certainly help, but even in the best case we’re looking at a few meters, as in electronic tollbooths.

Still, with a transmission range on the order of meters, a large-area sensor network composed of battery-free, passive sensors could be read using a drone, flying in a pattern over the network to gather the data.
Eric Yeatman’s group at Imperial College London has been developing the hardware platform needed for such drone-based data collection. Drones would navigate to each sensor-node location, power up the node, then collect data. To provide ample power, the sensor network incorporates supercapacitors that charge up via inductive wireless power transfer. Drones would work best for sensor networks having clear air space, for example, those on farms, aqueducts, pipelines, bridges, or dams.

Top shows a tremendous black cloud of smoke edged with orange and yellow over houses and trees. Bottom is a screenshot of a partial California map dotted with purple, red, orange, and yellow circles.
In November 2018, the Camp Fire, burning in California’s Butte County, sent thick clouds of smoke [top] into the San Francisco Bay area, where a network of sensors monitored by PurpleAir identified dangerous levels of airborne particulates [bottom]. The fire ultimately covered more than 150,000 acres (60,000 hectares), destroying 18,000 structures and claiming at least 85 lives.Top: David Little/The Mercury News/Getty Images; Bottom: PurpleAir

A large-area sensor network would have been very useful in managing the
Oroville Dam in California in February 2017, when a controlled release of excess rainwater caused the dam’s spillway to fail. The resulting cascade of water eroded the dam’s foundation, potentially compromising the dam’s integrity. Local authorities ordered more than 180,000 nearby residents to leave until more detailed inspections could determine that the dam was safe. Had a large-area structural-monitoring sensor network been in place at the time, those authorities could have gathered data to determine the state of the dam and make a timely and informed decision on whether evacuation would truly be needed. (Ultimately, the feared collapse did not occur.)

the 2018 Morandi bridge collapse in Genoa, Italy, was caused by a combination of aging infrastructure and severe weather. The disaster, which resulted in 43 deaths, might have been prevented if the weakening of the span could have been detected in good time by an installed sensor network, instead of by sporadic and sparse inspections.

Are event-driven or zero-power sensors ready to detect the outbreak of a wildfire in a remote area? We’re not quite there yet, but we are getting closer. All the essential pieces of such a large-area sensor network exist in various states of technical maturity; several more years of development and product integration will bring them to reality. Perhaps the more difficult challenge will be to motivate regional and federal governments to purchase and deploy such networks where they can be most useful or to enable a crowd-sourced sensor network, similar to PurpleAir.

Getting to zero-power sensors is well worth the effort and expense; deploying them to warn of wildfires would alone justify the R&D investment. Wildfires have already caused such huge losses and continue to threaten lives, property, habitat, and the long-term health of the millions breathing in smoke.

Imagine a future fire season in California. A lightning strike sets a tree ablaze, far from any houses, and the fire grows. But long before even a faint smell of smoke can wake your dog, the sensors in the forest wake up and alert a fire-monitoring station. At last, there is enough time and information to model the development of the fire, and to issue early evacuation warnings to the phones of everyone in the fire’s path.