Machine Learning-Based Modeling of pH-Sensitive Silicon Nanowire (SiNW) for Ion Sensitive Field Effect Transistor (ISFET)

Abstract

The development of ion-sensitive field-effect transistor (ISFET) sensors based on silicon nanowires (SiNW) has recently seen significant progress, due to their many advantages such as compact size, low cost, robustness and real-time portability. However, little work has been done to predict the performance of SiNW-ISFET sensors. The present study focuses on predicting the performance of the silicon nanowire (SiNW)-based ISFET sensor using four machine learning techniques, namely multilayer perceptron (MLP), nonlinear regression (NLR), support vector regression (SVR) and extra tree regression (ETR). The proposed ML algorithms are trained and validated using experimental measurements of the SiNW-ISFET sensor. The results obtained show a better predictive ability of extra tree regression (ETR) compared to other techniques, with a low RMSE of 1 × 10^-3 mA and an R² value of 0.9999725. This prediction study corrects the problems associated with SiNW -ISFET sensors.

Keywords: extra trees regression (ETR); machine learning (ML); multi-layer perceptron (MLP); nonlinear regression (NLR); silicon nanowire ion-sensitive field effect transistor SiNW-ISFET; support vector regression (SVR).

Figure 1

The (ML) techniques used for the Ids prediction of the silicon nanowires ISFET.

Figure 2

The Multi-Layer Perception Neural Networks (MLP).

Figure 3

Program flow chart of Multilayer Perception (MLP).

Figure 4

Architecture of Support Vector Regression (SVR).

Figure 5

Program flow chart of the Support Vector Regression (SVR).

Figure 6

Program flow chart of the Nonlinear Regression (NLR).

Figure 7

Program flow chart of the Extra Trees regression (ETR).

Figure 8

Comparison of SiNW-ISFET characteristics with varying wire lengths, obtained through four machine learning techniques and validated by measurements (Vds = 1 V): (a) SVR, (b) MLP, (c) NLR and (d) ETR.

Figure 9

Prediction accuracy of four machine learning techniques with different lengths of nanowires: (a) SVR, (b) NLR, (c) MLP and (d) ETR.

Figure 10

Comparison of the SiNW-ISFET characteristics with different numbers of wires, obtained through four machine learning techniques and validated by measurements (the nanowire length is 2 μm and Vds = 1 V): (a) SVR, (b) MLP, (c) NLR and (d) ETR.

Figure 11

Prediction accuracy of four machine learning techniques with different numbers of wires: (a) SVR, (b) NLR, (c) MLP and (d) ETR.

Figure 12

Comparison of the SiNW-ISFET characteristics with different gate lengths (0.73 μm and 3.73 μm), obtained through four machine learning techniques and validated by measurements (the nanowire length is 10 μm and Vds = 1 V): (a) SVR, (b) MLP, (c) NLR and (d) ETR.

Figure 13

Prediction accuracy of four machine learning techniques with different gate lengths: (a) SVR, (b) NLR, (c) MLP and (d) ETR.

Figure 14

Comparison of the SiNW-ISFET characteristics with different pH, obtained through four machine learning techniques and validated by measurements (Vds = 1 V): (a) SVR, (b) MLP, (c) NLR and (d) ETR.

Figure 15

Prediction accuracy of four machine learning techniques with different pH values: (a) SVR, (b) NLR, (c) MLP and (d) ETR.

Figure 16

Comparison of the sensitivities of SiNW-ISFET sensor, obtained through four machine learning techniques and validated by measurements (Ids = 200 µA and Vds = 1 V): (a) SVR, (b) MLP, (c) NLR and (d) ETR.