An artificial neural network approach for classification and prediction of low pressure chemical vapor deposition SixNy ISFET pH sensor drift

• Main Author: Mohammad Iqwhanus Syaffa, Amir
• Format: Thesis
• Language: English
• Published: 2019
• Subjects: TK Electrical engineering. Electronics Nuclear engineering
• Online Access: http://umpir.ump.edu.my/id/eprint/30027/1/An%20artificial%20neural%20network%20approach%20for%20classification%20and%20prediction%20of%20low%20pressure%20chemical%20vapor.wm.pdf; http://umpir.ump.edu.my/id/eprint/30027/

A sensor in an ion selective field-effect transistor (ISFET) is a field sensor in which ions within a sample media undergo multiple environments affecting reactions occurring molecules to concentrate at the gate oxide layer. A change in electrical charge occurs and it affects the conductance in the ISFET channels. Consequently, the change in conductance within the source and the drain produce an electrical signal. The most common problem is the occurrence of drift when the electrical signal output gradually changes independent of the measured sample. The primary objective of the present study is to investigate a reliable model based on artificial neural network to classify and forecast errors as well as to implement the drift compensation in low-pressure chemical vapour deposition SixNy ISFET pH sensor. Such model could be used to encounter voltage drift problems that usually exist in chemical sensors. Three units of ISFET sensors were used to calibrate with three types of pH buffer solutions that are pH 4, pH 7 and pH 10. Artificial neural networks were applied to construct a black-box with multiple inputs and outputs models of the ISFET data. A percentage accuracy value was used to assess the model’s performances in terms of classification, while the mean squared error (MSE) and the coefficient of determination (R2) parameter were used to determine the best models in predicting errors in the ISFET sensors. On the model’s structure in classification, Pattern Recognition Neural Network (PATTERNNET) proved to perform better than Function Fitting (FITNET) networks with 100% accuracy. The network configuration in PATTERNNET, a dual-layered network with 30 nodes on the first hidden layer and 3 nodes on the second hidden layer achieved the best results. As for prediction, the NARXBR model with 75 delays produced an efficient model in predicting errors in ISFET data set. The value of MSE = 4.8814e-5 and R2 = 0.99930 for the NARX-BR model showed that the model is capable of predicting errors. The drift compensation was applied and the drift issues in the ISFET sensors were successfully resolved. The present study demonstrates significant potentials in the development of artificial neural networks to stave off voltage drift issues in ISFET low-pressure chemical vapour deposition SixNy ISFET pH sensor.