Temperature Modulated MEMS Metal Oxide Gas Sensors

Tu Xiang Zheng

Metal oxide sensors are very popular as a consequence of their reasonable price and good durability.  However, they are lack of selectivity and response drift, which is why are used in low cost alarm-level gas monitors for domestic and industrial applications. It is important to propose new methods, which are able to improve the state-of-the-art in gas sensing. In order to do so, the temperature modulation of sensors has been proposed.

From electron spin resonance (ESR) measurements it has been proposed that adsorbed oxygen can be present in various chemical species transferring electrons from gas sensing oxide to the chemisorbed oxygen according to the following process:

O₂(gas)⇔O₂(ad)⇔O₂⁻⇔O⁻(ad)⇔O₂⁻(ad)⇔O₂⁻(lattice)

The temperature dependence of the different species has been examined. It states that an oxygen transition temperature is at 150 °C. Below 150 °C oxygen is mainly present as O₂^- and above chemisorbed oxygen in the forms of O^- or O^2- is present. This change in chemistry was correlated to a decrease in sample conductivity that occurred at around 160 °C. From these dependences the following basic mechanism for detection of combustible gases seems plausible: If the gas sensor is operated under ambient conditions it can be assumed that chemically adsorbed oxygen species are present at the surface. Combustible gases may react with these oxygen species and thus result in depletion of charged surface oxygen which in turn increases the conductivity of the gas sensing material. In this way combustible gases may not directly interact with the gas sensing material but its presence controls the concentration of pre-adsorbed oxygen, which controls the surface charge and thus the conductivity of the gas sensor.

According to this mechanism, it can be concluded that the modulation of a metal oxide sensor working temperature alters the kinetics of adsorption and reaction that occur at the sensor surface in the presence of atmospheric oxygen and other reducing or oxidizing species. It is reasonable to inference that sensor response patterns are characteristic of the species present in the gas mixture. Actually, many works have demonstrated that modulating the operating temperature of the sensors can achieve a high degree of selectivity. As an example, two components in a mixture of CO and NO₂ in air has been  simultaneously and accurately quantified by processing the response dynamics of a single micromachined tin oxide sensor operated in a temperature-modulated mode.

Similar to the above example, the MEMS metal oxide gas sensors proposed by the present author have their operating temperature modulated in a more efficient way. As well known, the thermal time constant of screen-printed sensors is quite large. As a result, up to now the temperature modulation frequency (20 MHz) has been too low and the corresponding principle-related response time (50 s) has been too high for many applications. With a special design, the thermal response of the MEMS metal oxide gas sensors is as low as 0.8 ms, as shown in the above figure. It compares much favorably with the thermal response of seconds found in conventional sensors.

The MEMS metal oxide gas sensor is based on a silicon wafer and fabricated utilizing CMOS technologies. Since the sensor is required to be operated at an elevated temperature a thermal insulating base is formed in the silicon wafer which is used to support the sensor body. Both a resistor for heating and a thermopile for temperature sensing are formed on the thermal insulating pad. Then depositing an electrical insulating layer and laying a tin dioxide layer is formed thereon. By employing such device structure with good thermal insulation to the silicon wafer, the sensor presents a series of advantages such as miniaturized size, low power consumption, and fast response.

In operation, the MEMS metal oxide sensor is exposed to a gas mixture, using a fully automated test setup, which consisted of computer driven mass flow controllers, a sensor chamber, and a data acquisition system for measurements in the millisecond range. The temperature of the sensor is varied by applying a modulate voltage to the heating resistor. Temperature range and frequency have been optimized.

Important features can be extracted from the sensor responses in two ways: the fast Fourier transform (FFT) and the discrete wavelet transform (DWT). Principal component regression (PCR), partial least squares (PLS), and multilayer perception neural networks (MLP) can be used to build quantitative predictive models. Then the different components of the mixture can be quantified precisely.