Exposing Pollutants With GaN HEMTs
The consequences of air pollution are alarming. According to the World Health Organization (WHO), each year a staggering 3.7 million loose their lives to outdoor air pollution, of which 80 percent die from a heart disease or a stroke, and the remainder have fatal respiratory illnesses and lung cancer.
Two pollutants are to blame for many of these deaths: dangerous airborne particles with diameters of several microns or more and NO2. Long-term exposure to the latter has been linked to bronchitis in asthmatic children, and it is known to impair lung growth in concentrations currently found in cities of Europe and North America. Another downside of NO2 is that it is the primary source of nitrate aerosols, which form an important fraction of dangerous airborne particles with diameters of 2.5 µm or less.
To curb deaths, the WHO has set an annual mean guideline for exposure to NO2 of 40 µg/m3, which equates to 20 parts-per-billion. However, this recommended upper limit is often exceeded, particularly in many of the urban centres throughout Europe. In London, for example, in just the first eight days of 2016, the far higher legal limit of an hourly mean of more than 200 µg/m3 of NO2 was breached on 18 occasions. One of the primary causes of these high levels of NO2 is the diesel engine, which may generate more of this oxide than one would expect, given the revelations emerging from the Volkswagen scandal.
In several big European cities, air quality monitoring networks are in place to monitor urban background and roadside emissions. These measurements, obtained with highly accurate but expensive analyzers, are combined with computer models to assess concentrations in places where monitoring sites are absent. It would be better to deploy more monitors. And thanks to the advent of the Internet-of-Things this might happen, because there is a growing interest in the use of miniaturized, low-cost air quality sensors that could create denser wireless networks. Such sensors could provide valuable complementary information to the existing monitoring sites. Progress has already been made in this field. Low-cost miniaturized sensors are now on the market, typically based on electrochemical cells or metal-oxide based devices (see “State-of-the-art miniaturized sensors" for details).
However, their sensing performance is not as good as that of conventional analyzers, according to air quality field testing campaigns, such as those organized by the European Network on New Sensing Technologies for Air-Pollution Control and Environmental Sustainability (EuNetAir). Weaknesses of these sensors include limited sensitivity, as well as cross-sensitivity to other gases, such as humidity and ozone.
More research has to be done to integrate these new technologies into smart sensing networks that can deal with some of their limitations. At the same time, demand for sensors that are selective, sensitive, and draw ultra-low-power is driving the research and development of a next-generation of sensors, known as microsensors and nanosensors. At the European microelectronics centre imec we are involved in this effort through the development of a novel gas sensor platform. It is based on the suspended AlGaN/GaN membrane and can combine operation at ultra-low-power with high sensitivity to NO2 and low interference from humidity.
In London, air quality measurement stations show NO2 levels often exceed the recommended maximum level set by the World Health Organisation.
As many of the readers of this magazine know, the GaN material system is already used to churn out billions of chips every year. Manufacture of blue, GaN-based LEDs is a well-established global industry, while, in comparison, sales of GaN HEMTs are not as high. However, they are being used to make RF and power electronics devices, and they show great promise as a generic platform for developing high-performance sensing devices.
Figure 1. Operating at 250°C using an integrated heater, imec’s GaN HEMT can distinguish between various concentrations of NO2 in humidified cleaned air. Upon NO2 exposure, the resistance increases until reaching a concentration and temperature-dependent equilibrium between NO2 absorption and desorption. Between NO2 exposures, the sensor recovers to the baseline by thermally induced desorption.
Exploiting the electron gas
Our sensors exploit the sensitivity of the highly mobile two-dimensional electron gas (2DEG) formed at the AlGaN-GaN interface. Thanks to the high sensitivity of the induced 2DEG to changes in surface charge it is possible to have a direct electrical readout of surface interactions involving charged species.
A search through the in scientific literature reveals that the GaN HEMT has already been used to form a variety of gas, chemical and health-related sensors. Surface functionalization of the gate area has enabled the detection of: certain gases, such as hydrogen; polar vapours and liquids, such as alcohols; biomolecules, including antigens and DNA; and electrochemical detection of pH and other ions, such as mercury.
We have found that these devices can be incredibly sensitive to NO2. Their response, defined as the change in device resistance, normalized by the baseline resistance (the device resistance before exposure to NO2) can reach the low-parts-per-billion range through precise recessing of the AlGaN layer at the gate area. Thinning this ternary dramatically improves performance, with sensitivity increasing by nearly three orders of magnitude as AlGaN is trimmed from 25 nm to less than 10 nm. Note that recessing is a known technique in gated devices, where it is used to improve HEMT characteristics.
With our sensor, the response to NO2 exposure is an increase in signal until it reaches a concentration dependent steady-state (see Figure 1). At that point, the sensor is in a state of equilibrium between gas adsorption and desorption. Following exposure the sensor recovers to the baseline due to gas desorption from the surface.
Our sensors are capable of detecting differences in NO2 concentration as small as a single part-per-billion, thanks to the extremely low noise level of the buried 2DEG (see Figure 2). This level of sensitivity is comparable to that of state-of-the-art chemiluminescent analyzers.
Figure 2. imec’s sensor can resolve single parts-per-billion steps for NO2 concentrations between 11 ppb and 200 ppb. The device was operated at 250°C in humidified cleaned air. The steps in the relative response are well above the noise level, indicating a sensor resolution below 1 ppb in the low ppb range.
The response and recovery times for our detector are strongly governed by a thermally activated process. Due to this, there is an exponential decrease in response and recovery rates with temperature. When the sensor is at room temperature response times can be as long as 1 hour, while recovery is very limited; but when the sensor is heated to 300°C, the steady-state response times (t90) can be as fast as 1 minute for NO2 levels as low as parts per billion. Heating is a double-edged sword, however, as faster response and recovery times resulting from heating must be weighed against increased power consumption.
We tackle this weakness by exploiting a unique feature of GaN-on-silicon technology – it allows removal of the substrate via plasma etching. We fabricate devices on 200 mm GaN-on-silicon (111) wafers with a mature process flow that is based on that of a power HEMT. But instead of processing the usual gate structure, we fabricate a tungsten resistive heater around the gate area. This provides integrated heating. The final processing step, as shown in Figure 3, is to plasma etch the silicon substrate below the active area to create freestanding AlGaN/GaN membranes – so called micro hotplates.
Figure 3. Fabrication of imec’s NO2 sensor begins with growth of an AlGaN/GaN heterostructure on an 8-inch silicon (111) wafer. The tungsten heater surrounding the structure is processed first, followed by the source/drain contacts and contact leads. Recessing the sensing area between the source and drain strongly enhances gas sensitivity. The final processing step is the etching away of the silicon substrate below the active region.
With this design, the power needed for heating is slashed 100-fold. Thermal losses to the substrate are eliminated, and heating requirements are restricted to just that of the active area within the membrane. Additional power savings result from duty cycling, facilitated by the ultrafast heating and cooling of the thin membranes. This enables power consumption to be trimmed to just a few milliwatts (see Figure 4).
Figure 4. The NO2 sensor's power consumption has been slashed by introducing a membrane architecture.
Another strength of our design is its superior rejection of humidity-based interference compared to state-of-the-art, metal-oxide-based sensors. Measurements of NO2 at relative humidity levels of 20 percent, 50 percent and 80 percent show that this has an insignificant impact on the baseline (see Figure 5). The NO2 response, meanwhile, is shifted by less than 10 percent by these variations in humidity.
Figure 5. One of the key advantages of imec’s HEMT-based device, compared with a conductive metal-oxide sensor, is the small interference from humidity. The graph shows the calibrated NO2 response for humidity levels of 20 percent, 50 percent and 80 percent. The effect of humidity on the baseline is insignificant. Note that humidity competes with NO2 absorption, decreasing the NO2 sensitivity at higher levels of relative humidity.
The superiority of our sensors stems from their design. With metal-oxide-based sensors, conductivity takes place across grain boundaries at the surface; but with our devices, the transduction mechanism is notably different, due to the burying of the conductive 2DEG beneath the surface. Nevertheless, the 2DEG density is balanced by the surface charge, so it is influenced by interactions of electron accepting or reducing gases with surface states. For example, our sensor is sensitive to ammonia in the parts-per-million range, but the response is opposite to that of NO2 – that is, resistance falls rather than increases. That’s because ammonia acts as an electron donor, while NO2 is an electron acceptor.
A rigorous assessment of the capability of any sensor has to involve field trials. We have taken our devices to a parking garage at our High Tech campus, where we compared their response to that of a chemiluminescent analyser.
Results show that the sensitivity of our device compares very well (see Figure 6). Both sensors reveal peaks in NO2 concentration that correspond to greater traffic at the start and the end of the work day, and variations in the background NO2 concentration. Unfortunately, these trials have also uncovered limitations with our devices. After several days of testing, significant interference in our sensor response appeared, probably caused by the presence of an unknown gas or vapour. This should not come as a surprise, given the large variety of gaseous and vaporous compounds that can be produced by burnt and partially unburnt fuel.
Figure 6. Field trials provide a rigorous assessment of the capability of the sensor. Engineers at imec evaluated their HEMT-based device in a car park.
It is not easy to address this issue, because it is related to the chemical properties of the sensor surface. Solutions include tuning properties or resorting to a filter. Another option is surface functionalization, involving the addition of chemically active structures – they could be (metal) nanoparticles, polymer layers, self-assembled monolayers, or advanced structures such as metal-organic-frameworks. This modification can deliver an additional benefit, as it extends detection to other interesting gases, such as CO2 and H2 (see Figure 7). However, although surface functionalization enables high tunability of the sensor surface, its practical application is a matter of on-going research.
Figure 7. The high sensitivity of the 2DEG to surface charge can be employed for the detection of CO2, a gas that is currently gaining interest for indoor air-quality monitoring. In contrast to NO2, CO2 does not interact directly with the sensor surface: even exposure to 10 percent CO2 does not result in any response. However, after simply coating the surface with a pH sensitive polymer, such as polyethylenimine, concentrations well below 500 ppm can be detected. This is sufficient for indoor use. The sensor is able to detect CO2 because it induces pH changes within the coated layer. Specifically, the absorption of CO2 results in protonation of polymer side groups, which effectively gate the recessed 2DEG channel and increase device current. Note that detection of hydrogen concentrations in the ppm range is possible via thermal evaporation of a 5 nm-thick platinum layer on the recessed gate area.