Multi-modal sensing system is essential in healthcare and internet-of-things (IoT) [1]. Making the sensor systems small, low power and cost effective is the ultimate goal to fulfill the growing needs. To achieve this requirement, we present a single-device-dual-sensor by transforming a conventional dual-gate ion sensitive field-effect transistor (DG-ISFET) [2] to a pH/light bi-functional sensing device. In addition to pH sensing ability, since photocurrent in DG-ISFET has been proved to be controllable by gate voltages [3], it is possible to detect light by a DG-ISFET. However, the signals of pH and light are coupled together in a complicated manner. Therefore, in order to realize an effective dual-sensor, we introduce a sequential re-configuration method and back-propagation neural network (BPNN) to DG-ISFET in order to distinguish and quantify the signal from pH and light generated by the same device. The sequential re-configuration (SRC) method used in this work aims to mimic the function of sensor array in a multi-sensing system. To achieve this goal, the gate bias of DG-ISFET is intentionally configured in a sequential manner by tuning top and bottom gate voltages. In other words, we virtually generate a number of sensors with different sensitivities to light and pH from a single DG-ISFET device to replace the sensor array in the multi-sensor system. Afterwards, the output signals are processed by machine learning algorithms to quantify the actual value of the analytes. To quantify the pH value and light intensity, BPNN serving as a regression machine learning algorithm are employed. The regression results given in the form of boxplot of pH value and light intensities is shown in figure1. Based on the results, the error between calculated and true intensity is smaller than 10 μW/cm2 or 1.1% of every level of light intensity. pH values, on the other hand, have smaller calculation error of 0.5%. Based on these results, BPNN models are capable for practical applications to measure pH value and light intensity. Furthermore, the algorithm-calculated results have smaller variation than the direct-measured data does. It indicates the process variation of different sensors can be mitigated by the proposed SRC method with algorithms. This SRC based DG-ISFET reduces the need of sensor array and improves data consistency compared with a real sensor array. At the same time, back-propagation neural networks are employed to construct the data processing models to realize multi-signal quantification. Based on this work, the light interference of DG-ISFET can be transformed into effective illumination sensing quantity. And the proposed SRC-based single-chip-dual-sensing technique has good potential to be applied to sensor fusion technologies for healthcare and environmental monitoring.
Read full abstract