Environmental Sensors Catch More Data For A Greener World

Sensors to detect temperature, pressure, and gases, such as CO₂, have been around for centuries. However, the latest devices can measure a growing list of substances and process the data in real-time. Likewise, single-use sensors to measure pH levels in water are well established, but the latest water sensors can be deployed all along the pipeline from source to processing to outlet or tap, saving a long and costly trip to the lab. Some sensors are flexible and printed, and many are connected to onboard MCUs.

Together, these devices will deliver a ton of data to companies, researchers, and eventually maybe to consumers, who can use it to provide better quality water, air, and soil to support human and planetary health.

“Having high-quality sensors available and accessible to people makes things better, and it’s oftentimes in ways you don’t understand,” said David Fromm, COO at Promex, which develops specialized manufacturing processes to build customer widgets. “The power of crowd-sourced information is incredible. With things like air quality, air sensors, and temperature, people like to measure that stuff in their backyard and have that data. If I can make it an inexpensive sensor that people can buy, that data can go up into a pool and they can create these incredible heat maps of various phenomena, and that could be the same on a water system. Consumers, in this case, feel like they’re contributing. People find it interesting. Then the power of that data can be game-changing, because you’re getting this high fidelity, dynamic image of measurements at different locations. They’re not stationary. You’re getting everything at all times.”

General knowledge is power. “If the consumer wants it, and especially if they’re willing to pay to obtain that knowledge, let them,” said Fromm. “It’s incredible to see what people are starting to do and the ability to assimilate across applications, that people are getting very clever about what they’re doing. It’s not just the technology, but it’s how that technology is being applied to solve real problems. We get to see an incredible cross-section of that and try to help them make it real.”

Water sensors
The city of Flint, Michigan, can offer lessons about lead in the water supply, but it’s not alone. Studies in Missouri, where Brewer Science is based, found varying degrees of lead in the water supply of most schools. The lead concentration varied by tap, not by building, because of the local distribution.

“There are schools where every single water fountain or water source tap has elevated lead in it — 20, 40, 50 parts per billion,” said Adam Scotch, director of R&D for Smart Devices at Brewer Science. “There are some where it’s just one or two taps. It is such a local problem, and they didn’t know.”

Brewer has devised printable, flexible sensors to fill the gap between guesswork and expensive water testing systems. “We should be able to monitor the problem with water, and it’s similar with electricity,” Scotch said. “It’s all very regional. Every town has its own water supply system, or well system, or they have their own water treatment system. There is no formal infrastructure that everything is connected. We’re still using nineteenth-century pipes in some parts of the country — wood and lead — and that creates a problem. It’s not just about the sensors. The easy part was finding a way to detect the material, the sample. Anybody can do that in a lab. The hard part is turning that into a product that will last. How do you put down this sensor material? How do you put a connector on it? It’s all about inputs and outputs. The semiconductor industry is struggling with the same problems. It’s always, “How do you get your data from here to here, from your chip to your PCB, from your PCB to your hard drive, or whatever?’ It’s always about the transfer of that information.”

Fig. 1: A water sensor system that can be deployed anywhere along the water flow. Source: Brewer Science

Besides lead, many types of contaminants can end up in water and cause issues. For example, radon has been linked to lung cancer in non-smokers. “Dissolved gases in water, generally, are more difficult,” Scotch said. “Ammonia is a bigger problem than radon, because there are lots of sources of nitrogen — fertilizer and farm runoff that tend to put too much nitrogen in our water supply, and then that converts to nitrites and ammonia, and measuring that is tricky. Arsenic is another big one, because arsenic is naturally occurring in many well water systems.”

A PFAS sensor is under development. “It’s still very early stages, but it is also a real-time sensor, and it’s the first one that I’ve seen on the market,” said Scotch. “Right now it’s hard enough to measure PFAS in the fab. We can detect 10 to 20 parts per trillion, when the proposed EPA limit is down into the single-digit parts per trillion water.”

It serves everybody’s best interests to know what they are drinking. “What we’ve discovered is that there’s a huge need, a huge market pull — I’ve never seen a pull like this in my career — that people don’t know what the quality of their water is, or their water supply out of their taps,” said Scotch. “Industrial customers need to monitor their effluents. Everything is based around sampling and testing in a lab, either on-site or they have to send away samples. It’s costly and it’s only a single sample, at a single point in time.”

Singular data points can only help so much. The real value in understanding is rate of change, what’s changing, what does that delta look like, and why is it important. From there you can investigate why something has changed.

Near-field wireless technology sends the data in batches every 10 minutes or hour or so, which is a significant improvement from waiting months for labs results. “We started with cellular in the very, very early days when we were installing sensors in plants,” said Scotch. “These plants were vast, and we couldn’t create a node network to be able to communicate. But as low Earth orbit satellites have come online in recent years, there’s no reason why satellite communication couldn’t be an option, especially if you’re in the middle of nowhere, in a field.”

Air sensors
Researchers at Harvard University have been working on sensors that sniff like a dog, meaning they can take longer or shorter sniffs based on what they detect, mimicking a real nose. The project began about five years ago, funded by the U.S. Federal Railroad Administration. It focused on real-time detection in field sites for gas and oil tanks and derailments, and how you can judge flammability considerations and hazards for transportation goods.

“We investigated how sniffing sequences, which is bio-inspired by how canines sniff, can change and modulate how we intake the air, and how that influences dynamics and kinetics,” said Haritosh Patel, a graduate student at Harvard University’s School of Engineering and Applied Sciences. “There are a few takeaway messages from the work. One is that time is a very important aspect to chemical sensing, rather than the sensor itself having threshold readings, where you just compare the end state to the start state. We know that can cause a lot of false positives and negatives in detection, because the refractive index for this particular sensor would be similar across a lot of different chemicals. When we were just looking at that, we trained a machine learning model, and we had as low as 40% accuracy. But with the same hardware system, when data was collected over time — and where the absorption kinetics were not tossed away, but actually used as a means to make a prediction — it became evident that we can do classification between very similar chemicals, like pentane to decane, or some of the alkanes that are used in the oil industry.”

The team demonstrated how sniffing can quicken the process of detection — so you’re not waiting for passive diffusion of the molecules to the sensor, but you’re actively doing air sampling — and also differentiate chemicals where some would saturate earlier than others, based on this molecular weight, volatility, or other characteristics. “Active modulation of the air sampling, akin to sniffing, is an important concept in which we can actively use muscles to make the hardware see different things from the same concentration,” said Patel.

The sniffing is achieved through use of fans in the sensor. “The fans are laid out in a central flow stream so we can have it in a passive mode, where there’s no sniffing, and you just let the volatiles reach equilibrium in the sensor casing, and you see the response,” Patel explained. “But then we lose the picture of the non-equilibrium kinetics, which was the real key advantage we saw in earlier works. To gain back the non-equilibrium kinetics of adsorption desorption onto the sensors, sniffing is employed through use of fans — think about small mini fans that are right on a CPU device — and you use it to modulate the flow rate of incoming air sample, and resetting the sensors, but with an outgoing air sample. These fans are modulated in specific sequences now, which can tune if it’s an inhale of a 3-second sequence or inhale of a 30-second sequence, and the timescales of that are dependent on the chemical concentration and location of the sensor.”

Driving the AI/ML is an MCU. “Right now, the microcontroller is not the key aspect of the device,” said Patel. “The main goal of the microcontroller is to use a multiplexer to get [data] from both digital metal sensors and analog sensors. It synchronizes all the firmware to take the data sets in a specific frequency, whatever the resolution frequency we set, and then transduces that into the analog signal in the microcontroller. And finally, with a Bluetooth or Wi-Fi connection, the data is transmitted to a cloud system for a database, where all of the data sets are then stored for further machine learning training in the later iterations, or off-board computation, where a quick machine learning model can be consulted, with immediate feedback of exact concentrations relayed to a building manager.”

Managing air quality in commercial premises is a major application, but the sensors can be used to measure everything from toxic 3D printing art supplies in schools, to early mold detection in homes, food spoilage, or formaldehyde from new carpets and furniture. It could be used for tamper-proofing perfume and alcohol. First responders could have the technology in AR glasses to guide them through a safer chemical path in a hazardous site, and robots could have it to help detect humans under rubble.

Environmental sensors and machine learning
Other companies are also working in the sensor space to help keep the planet green.

“We’re very conscious of our environment,” said Tanja Hofner, lead principal systems hardware engineer at Infineon Technologies. “We see growth with environmental sensors — everything from humidity to pressure to gas sensors. You see a lot of radar sensors, where you would want to control how often the HVAC turns on.”

Key concerns include catching anything coming from a battery, such as a fuel cell in hydrogen power, or CO₂ gas produced through an evaporator coil in HVAC. “CO₂ gas sensors are a big thing right now,” said Hofner. “There are different technologies for it. We have technology, which is a thermal conductivity sensor that creates a controlled environment by heating up the gas chamber the unknown gas comes in, and then compares the values to a hermetically sealed gas chamber that contains nitrogen.”

The onboard chamber is used as a control study for anything detected outside. “They are connected to an MCU, or M04 controller that runs the intelligence on board,” Hofner explained. “It runs an algorithm that compensates for any kind of environmental challenges, such as humidity or temperature change, because gases are very susceptible to any of those.”

The Covid-19 pandemic illustrated the importance of monitoring air quality. “It was very important that we could track gases in rooms, in schools, in public places,” said Hofner. “Then we could say, ‘We need to up the ventilation, open the window, or the doors in order to make sure that everybody doesn’t breathe in the same viral contamination that everybody else is breathing out.'”

Like the Harvard sniffer, many of the latest sensors mimic the human senses. “We have touch sensors showing capacitive touch and inductive touch sensing,” said Hariharan Mani, semiconductor applications engineering leader at Infineon. “We can do liquid-level sensing with our cap touch sensors, as well. We have gas sensors, which are mostly remote for industrial and sometimes consumer applications. We have our microphones, which represent the audio hearing portion of it. We have vibration sensors that can be used in combination with audio. You can do predictive maintenance, you can do voice communication, you can do active noise canceling. We also have radar sensors, which represent the vision aspect without the privacy concerns of a camera. We’re trying to mimic the human senses so that our devices, which are intelligent, can be contextually aware. The next step is digitalization of the sensors and getting useful information.”

For example, an ML-based processor can incorporate very advanced graphics on the output side after integrating with a lot of sensors on the input side. Each sensor could be used separately for a single application, or they could be combined in robotics, smart thermostats, or security cameras. “All of these kinds of applications require multiple types of these components,” said Mani. “It might not require everything. For example, CO₂ is a very special use case device, but audio and radar and PSoC Edge with cap sense — all of these can go hand-in-hand in a lot of different applications.”

Sensor devices frequently are connected to edge AI MCUs, with CPUs, NPUs, or GPUs. Which processor type often depends on the size of the end device, explained John Weil, vice president and general manager for IoT and edge AI processors at Synaptics. For example, a 3cm device containing a 12mm x 13mm MCU is an easy option, though smaller is possible.

“The devices are very doable,” said Weil. “They can run on AA batteries, and by putting something in a small form factor, it can be used to test water quality or air quality. These products can do that. That’s not an issue at all, and there’s more than enough compute to do it. It’s really a function of the sensor. We’re limited only by the sensor quality today, not by the compute at whatever size.”

A lot of world-class sensor capability is still not in production. “It’s been in design, R&D, back-of-a-lab kind of stuff for years,” said Weil. “I’ve seen sensors that could predict the difference between a French Bordeaux and an Italian wine. That technology exists, but you can’t buy it. We’re still a few years away from the researchers being able to really get it productized.”

Reliability and predictive maintenance
Certain types of sensors have struggled to complete a repeat performance, meaning that reliability is a key hold-up in advancing the technology.

“I can get it to work one time. I can get it to work 100 times,” said Weil. “Can I get it to work more than that? It’s a chemistry problem, because you can make test strips for water, for salinity, for all these different things. It’s fire-and-forget one-time use. But how do you make a sensor that can be immersed in whatever it is, gas, liquid, over and over and over, and not have a degradation or a reliability issue?”

Brewer’s water sensor can last for six months in various harsh conditions, while Infineon’s CO₂ sensor is geared towards a lifetime of 15 years, typically in an industrial application.

“Usually when you are looking at an HVAC system or a fuel cell system, or any kind of system that tracks these kinds of gases, you would look at a limited lifetime there, as well, which is the lifetime of the sensor,” Hofner said.

One of the final frontiers in making these kinds of devices is making them smart enough to know when they’re behaving correctly or not. “When you’re deploying something in the field, you can’t just have an army of service techs dealing with them,” said Promex’s Fromm. “The economics don’t work. They have to be smarter. So edge AI — that’s the future in the next five years, for sure.”

This is where predictive maintenance fits in. Sensors must be able to send alerts before a system completely fails, or before an entire line gets taken out. “It can do it in such a way that you don’t have to send people in on a regular schedule,” said Pallavi Sharma, director of product management at Imagination Technologies. “If you still risk missing something, you can run more detailed diagnostics to make sure that you’re catching even the nuances and running any kind of updates or optimizations in time. If you can detect that a certain portion of this is not working, that some sensors aren’t reporting correctly, you can just repair or replace those little components.”

The chips are different, depending on whether they are used for real-time alerts or preventive maintenance sensors. “The OS is different. The loads are different,” said Sathishkumar Balasubramanian, head of products at Siemens EDA. “You need to have a chip that has adaptability to a sensor and everything else. The package is different. What kind of software goes into it is different, and what kind of code. You don’t need a big process of code, because you’re not solving a bigger problem. You’re essentially taking an input and making some decisions, and following and transmitting the data back into the central server.”

Conclusion
Sensors have come a long way but there are untold possibilities of where they might go next, and there’s much more to do. “We just have to be mindful that we can’t solve everything in a day,” said Brewer Science’s Scotch. “It takes time. But then also, what needs to be measured? Every sensor requires its own tweaking of the chemistry.”

Another challenge is that once people have the information that something is wrong with their body, water, or another environmental metric, they may not have the ability to do anything about it. For example, the Missouri water study was unfunded, which hampered the next steps. “If they measure an elevated amount of lead, they have to either shut off the tap or do remediation, and they don’t have the money for that, either,” said Scotch. “Unfortunately, we see this not just in the school districts. Some people don’t want to measure their water because they’re afraid that they’re going to have to take action. If they don’t measure it, then they don’t know. This is a challenge. This is going to be a dangerous area in the future. It’s a good time to be a plumber.”

When people get sensors in their hands, they will start appealing to authorities to fix things.

“It’s necessary, because we’ve abandoned our water, which is our most precious natural resource,” said Scotch. “We’re always worried about oil and coal running out. There’s always concern about fossil fuels in general being depleted, or mining of precious metals, but nobody really thinks about water in quite the same way, until more recently.”

Likewise, it is only the beginning for air quality detection. “One of the things that helped the discovery and invention of what we call universal smellers, is this notion that we were hyper-fixated on the hardware being the key and only role to get there,” said Harvard’s Patel. “Biology indicates that the anatomy of your nose matters, how fluid transports through the nose, all the chemical structures of the nose, the way the brain even processes it. There’s more than just hardware. There are all these aspects, and learning about each of these aspects in its role in the final outcome, which is chemical detection, can aid in making applications. Some applications need to use some of these strategies, and not all of them.”

Materials, too, are important. “No doubt they have a very critical role in it, but we’ve studied enough materials where we have reached the sensitivities we need,” said Patel. “Now we need to use the other parts of the strategies to get towards universal sensing. That’s a very exciting thing to capture, where machine learning works directly with hardware.”

— Ed Sperling and Gregory Haley contributed to this report.

Related Reading
Wearable Connectivity, AI Enable New Use Cases
New types of wearables and devices can record bodily data or simulate the senses without needing to meet stringent med-tech rules.
Med Tech Morphs Into Consumer Wearables
Smart watches, rings, and a growing array of patches are adding more functionality and being used across a growing set of applications.
Precision Under Pressure: Managing Materials Complexity In Advanced Packaging
As packaging integrates diverse substrates, adhesives, and exotic metals, manufacturers must rethink precision from the ground up.
Advanced Packaging Depends On Materials And Co-Design
New materials play a pivotal role, but solving integration problems remains a challenge.
Upcoming Challenges And Changes In Semiconductor Materials
Brewer Science’s new co-CEOs talk about shifts in materials for IC manufacturing, flexible substrates, and the uncertainties in advanced packaging.