Please use this identifier to cite or link to this item: https://ptsldigital.ukm.my/jspui/handle/123456789/515012
Title: Ion selectivity studies using supervised neural network for ion-sensitive field-effect transistor sensor
Authors: Wan Fazlida Hanim Abdullah (P39409)
Supervisor: Mohd Alauddin Mohd Ali, Professor Dr
Keywords: Ion selectivity studies
Neural network
Ion-sensitive field-effect
Transistor sensor
Field-effect transistors
Issue Date: 5-Aug-2011
Description: The ability of field-effect transistors to act as an electrochemical sensor opens the door of integrated circuit technology mass-production benefits to chemical sensing applications. In view of fulfilling environmental and health monitoring sensing requirements, one of the challenges for the ion-sensitive field-effect transistor (ISFET) sensor is the need to demonstrate high selectivity from the angle of accuracy. The central issue in this thesis is the study of ion interference and improving ISFET signals selectivity electronically. The research work focuses on sensor signal interpretation of main ion concentration in the presence of interfering ions, particularly potassium ion (K+) in interfering ammonium ions (NH4 +), with supervised learning neural network post-processing based on primary data. The research area is multidisciplinary, covering device characterization to circuit design and electrochemistry to mathematical algorithm development. Designing of interface circuit was targeted for K+ and NH4 + ISFET array parallel measurements with constant-current constantvoltage transistor biasing utilizing a single reference electrode. Training data collection was aimed towards a 4-sensor array response in the range 10-6 to 10-1 mol/L molar concentration samples. Sensor voltage response was acquired to act as input data while prepared sample concentrations from standard calculations was used as target for training data. Multilayer perceptron architecture with back-propagation training algorithm was designed in Matlab software environment keeping the modularity of addition/multiplication in view of future silicon implementation. As a neural network performance improvisation technique, an ensemble system was constructed using bootstrap aggregating approach with the output combined by averaging for regression and voting for classification. Results based on semiconductor device transfer characteristics clearly showed that interfering ionic activity influenced the threshold voltage and lead to reducing the main ion detection range. To overcome erratic sensor behaviour that was captured to have low repeatability and large mean squared error, sensor voltage signal was referenced to sensor response in deionized water prior to measurement. This significantly improves the repeatability correlation factor by 15.5% and reduces mean-square error by 98.3%. Multilayer perceptron feedforward neural network with single hidden layer was able to estimate test data with 15% improvement over direct estimation without neural network post-processing. Findings indicated that averaging and voting by multiple classifiers improved performance by a further 5%. Furthermore, ensemble system was able to avoid unpredictable exceptionally weak estimations in regression by a single classifier thus improving performance stability. In conclusion, the thesis findings corroborate the implementation of neural network post-processing towards improving the accuracy of device sensor reading interpretation. In particular, the thesis provides clear evidence that referencing the voltage signal to sensor response in deionized water is able to improve quality of training data for supervised learning. Additionally, the thesis supports the seeking of opinion from ensemble classifiers towards more stable ion concentration estimations.,Ph.D
Pages: 120
Call Number: TK7871.95.W3518 2011
Publisher: UKM, Bangi
Appears in Collections:Institute of Microengineering and Nanoelectronics / Institut Kejuruteraan Mikro dan Nanoelektronik (IMEN)

Files in This Item:
File Description SizeFormat 
ukmvital_73874+Source01+Source010.PDF
  Restricted Access
7.08 MBAdobe PDFThumbnail
View/Open


Items in DSpace are protected by copyright, with all rights reserved, unless otherwise indicated.