Integrated biosensor platform based on graphene transistor arrays for real-time high-accuracy ion sensing

Abstract

Smart sensors, such as sweat sensors, targeting various physiologically relevant biomarkers in biofluids have recently demonstrated great potential for health-tracking and medical diagnosis. Two-dimensional (2D) materials are particularly promising in biochemical sensing applications thanks to their large-surface-to-volume ratio, which allows the sensor channel to be readily modulated upon chemical changes near the surface, translating chemical signals into the electrical or optical domain with enhanced sensitivity. Graphene is the most widely studied material among all 2D materials. It has the largest surface-to-volume ratio and exhibits a number of promising characteristics. Despite advances in material synthesis, large-area integration of devices based on graphene and other novel materials still suffer from strong device-to-device variability caused by intrinsic defects, gate oxide nonuniformities, and parasitic effects. Device fabrication also introduces additional variability in sensor response from batch to batch. Here we demonstrate an approach to overcome the challenges in 2D material-based sensing devices and achieve high performance and enhanced functionality. Rather than focusing on the improvement of intrinsic material quality, fabrication uniformity and surface functionalization, we develop a high-density graphene-based sensor array platform to overcome the large degree of variability of advanced materials. We fabricate arrays (16×16) of graphene devices to provide more than 200 working sensing units for each chip, and configure them into multi-ion sensors by functionalizing the surface with three different ion-selective membranes (ISMs). Ionized calcium, potassium, and sodium were chosen as analytes of interest due to their commonplace in diagnostic tests and their physiological importance in blood, urine, and sweat. We demonstrate near-ideal sensitivity, excellent reversibility, and large detection range for each type of sensor despite non-uniformity in individual devices. The variations and imperfections in material synthesis and device fabrication can be leveraged by statistical analysis and machine learning algorithms. A profile-matching calibration method utilizing sensor non-uniformity and redundancy is introduced to eliminate the need for multiple calibration solutions, which is especially useful for sensing applications targeting portability and field use. A Random Forest algorithm is used to quantify analyte concentrations in the presence of multiple-ions. The abundance (N>200) and multiplexity of sensors and sensor types are shown beneficial for improving model accuracy. We demonstrate that system-level co-design of sensing arrays and algorithms significantly improves sensor performance thus enabling rapid prototyping and in-depth data analysis in spite of the limitations present in graphene and other advanced 2D materials.