While aggressive transistor scaling has been at the forefront of development in the semiconductor industry for decades, there is currently an increasing demand for added functionalities in a single device. This is most evident in the rapid development of integrated smart sensors in many industries (smart homes, automotive, health and medicine, etc.) and the rise of wireless sensor networks and the Internet of Things. Because initial integration attempts through system-on-package suffers from long wires and high RC delay, the pinnacle of sensor integration lies in its fabrication within a CMOS processing sequence on a semiconductor substrate and three-dimensional integration (Fig. 1). Through the integration with CMOS circuitry, the processing of the read signal can be performed quickly and efficiently, in terms of power dissipation.
The CMOS image sensor is a perfect example of the revolutionizing effects this type of development can have on a research field and in industry. More recently, gas sensors, which used to be very bulky and power demanding, have entered the realm of semiconductor integration, resulting in their implementation in hand-held devices. Currently, there are three main principles upon which solid-stage gas sensors are in large scale use: electrochemical sensors based on solid electrolytes, pellistors based on catalytic combustion, and semiconducting metal oxide (SMO) sensors based on conductometric or chemiresistive sensing. The SMO has the advantage of compatibility with CMOS fabrication, which is why its use has prospered. Many other research activities for a variety of sensor technologies is ongoing including building automation, smart traffic, biosensing, spintronic materials, temperature sensors, and many others.
The added benefits of SMO sensors is their high sensitivity to many polluting and toxic gases (VOCs, CO, CO2, CH3OH, NH3, H2S, and many more) at very low concentrations, with fast response times, ease of maintenance, low cost, and scalability towards their use in portable devices. One downside of SMO sensors is their poor selectivity towards a target gas. Since the SMO depends on reactions between gas molecules and its exposed surface to cause a change in the film resistance, it is very difficult to recognize exactly which molecule has adsorbed. One method is to manipulate the fact that the adsorption, or ionosorption, of different gas molecules varies depending on the applied temperature, which must be between 250°C and 500°C for proper sensor operation. Therefore, selectivity can be achieved using a sensor array, where each sensor is heated to a different temperature and the combination of all sensor outputs can be used to better identify the gas which has adsorbed.
A sensor array requires many sensors, each with their own microheater, which can drive the operating power of these devices to very high levels. One of our achievements is the design and simulation of a microheater array (Fig. 2), which is extremely useful for selective low-power SMO gas sensor integration. The microheater array allows us to simultaneously provide different temperatures to small SMO layers. The sample layout from Fig. 2 can provide 270°C and 350°C at the same time, giving more information about the gas molecules which might be adsorbed on the SMO surface (Fig. 3). This design provides a significant reduction in power, operating at below 10mW during a sensing cycle, opening up the potential of its use in portable and wearable electronics.
A key concern for these types of sensors is thermal management, because temperature uniformity affects sensing quality, while thermal dissipation from the microheater is the primary source of power losses. For proper thermal insulation, three heat loss mechanisms (Fig. 4) must be analyzed:
The heat losses, and the relationship between the applied power and achieved temperature, can be modeled using different techniques, including analytical equations, a Cauer network, or with finite element model (FEM) simulations. Although FEM is the most accurate, it is also the most cumbersome to set up and time consuming. The analytical equations can only provide a rough estimate, while a Cauer network allows us to discretize the entire sensor geometry to find a balance between accuracy and speed. The thermal parameters, which define the thermal behavior of the sensor, are mapped to an electrical circuit equivalent, allowing for simulations using common circuit simulators, such as SPICE. Each discretized component can be represented as a parallel combination of a thermal capacitance, thermal conductive resistance, and a convection resistance, driven by a heat flow, and represented by an equivalent electric current. This way we can model a complex geometry using a circuit simulator in order to find the influence of the geometry and different materials on the power dissipation and temperature distribution on the active sensor area. Ultimately, this information helps to optimize the design and find more appropriate materials for the integrated sensor structure. A comprehensive electro-thermo-mechanical approach must be undertaken to analyze and design integrated sensor structures.