Quantitative nanoscale field effect sensors

Abstract

Semiconductor nanowire field effect transistors have emerged as a promising technology for development of label-free biomolecular sensors for rapid diagnostics. However, their practical application has been hindered due to the significant device-to-device variations in electrical properties and the need for individual sensor calibration. Recent advances in device fabrication and demonstrations of multiplexed sensing and quantification might make this technology more competitive with respect to the current cutting-edge techniques such as surface plasmon resonance.

Keywords:

• nanoscale
• biofet
• isfet
• label-free
• calibration

1. Introduction

Rapid and reliable detection of biomolecules using direct electronic label-free detection in disparate environments has become a subject of great interest for biomedical and clinical research. Significant research and industrial efforts are directed towards development of point-of-care (POC) diagnostic tools which could provide cheap and fast analysis at the patient's location. Semiconducting nanowires configured as field effect transistors (FETs) have demonstrated the promise to deliver low-cost portable electronic platforms for rapid, ultrasensitive and multiplexed detection of biomolecular species. Compared to the current cutting-edge techniques such as surface plasmon resonance or DNA microarrays, direct electronic detection allows integration of FET based sensors with data processing components such as registers, operational amplifiers, analog-to-digital converters, etc.

The application of FETs for biomolecular detection was first introduced three decades ago [1]. An ion-sensitive FET (ISFET) [2] is similar to a conventional metal-oxide-semiconductor FET (MOSFET); however, instead of a metal gate that is used to modulate device current, an ISFET lacks the metal gate, thus allowing the oxide layer to be exposed to the electrolyte solution. Modulation of the surface charge changes the surface potential of the sensor thus causing an increase or decrease in device current. In addition, binding of charged biomolecules (proteins, oligonucleotides, etc.) from the solution to the ISFET surface can also cause changes in the semiconductor surface potential.

Nanowire FETs were first introduced in 2001 by Lieber's group [3] and since then they have demonstrated various application such as detection of antigens, oligonucleotides, viruses and cellular functions. However, not much data were demonstrated towards the repeatability and reproducibility. Furthermore, one of the major shortcomings of the technology was the inability to produce devices with low electrical variations which made sensor calibration and quantitative sensing practically impossible. Most of the work done by 2007 was based on the bottom-up or chemical vapour deposition (CVD) grown nanowires [4]. Significant variations in electrical characteristics [5] and difficulties to incorporate CVD grown nanowires in existing top-down fabrication techniques [6] represented major shortcomings of this technology.

Improvements in the quality of silicon-on-insulator wafers [7] and processing techniques [8] gave a significant boost to the use of top-down fabrication techniques. Using top-down fabrication techniques, a nanoscale device is formed by either electron beam lithography (EBL) or by deep ultraviolet (DUV) lithography, followed by reactive ion etching (RIE) [9]. Recent experimental studies [10] on 1/f noise of the RIE defined devices have demonstrated significant degradation of nanowire electrical and transport characteristics such as carrier mobility and sub-threshold swing as well as an increased number of traps.To date a major drawback of the bioFET applications is considered to be the lack of calibration methods and inability to quantify biomolecular species. More importantly issues such as repeatability and reproducibility of label-free detection have not been addressed. This problem has prevented bioFET technology to become competitive with the current mainstream techniques, such as surface plasmon resonance [11] or enzyme linked immunosorbent assay (ELISA) [12].

Another drawback of the bioFET technology is the inability to sense in physiological environments such as whole blood or plasma. Negative side-effects such as bio-fouling, non-specific binding, false positive detection and most critically Debye screening limitations [13, 14] have diminished the competitiveness of the bioFET technology and have limited its applications to academic research.

However, recent papers have demonstrated several successful approaches which might solve above-mentioned issues. These methods successfully utilise the advancement of microfabrication techniques to introduce microfluidic purification chips which serve for blood filtration prior to sensing and are based on capture-release [15] and size-exclusion processes [16].

2. Fabrication

There are several drawbacks of using CVD-grown nanowires as FET based sensors and incorporating them in mass production techniques. First and foremost, the use of CVD grown nanowires is incompatible with mainstream complementary metal oxide semiconductor (CMOS) technologies which makes integration with on-chip amplifiers and data processing circuits practically impossible. The inability to integrate bottom-up nanowire FET with microelectronic components and microfluidics directly hinders application for multiplexed detection schemes. Secondly, the low yield and inability to produce large number of devices with similar electrical characteristics [5] hinders device calibration and does not allow analyte quantification which is necessary to stay competitive as POC diagnostic tool.

In the pioneering work, Stern et al. [17] introduced a CMOS compatible nanowire process using EBL with lateral dimension on the order of tens of nanometers. A critical step in this technique is anisotopic etching of (100) silicon using tetra methyl ammonium hydroxide (TMAH) which etches (111) planes 100 times slower than others.

The first step of this method assumes thinning of the active silicon layer using dry oxidation to achieve the desired thickness of the active area. Upon the optical definition of the nanowire body (mesa) and back gate via etch, implantation was performed to form source and drain contacts followed by activation annealing. Using dry oxide as a mask or hydrogen selesquioxane (HSQ, a negative electron beam photoresist) device, widths were scaled down from micrometer size to several tens of nanometers using EBL. Upon exposure to an electron beam, the HSQ forms an amorphous silicon oxide layer whose crystallinity can be further improved by oxygen annealing.

As mentioned before, the crucial step for scaling down nanowire dimensions is an anisotropic etch using TMAH. This allows precise control of device width as a function of etch time. Due to the orientation dependent etch, the cross section of the nanowires obtained this way is trapezoidal with a characteristic angle of 54.7° in the base (Figure 1).

Figure 1. Scanning electron micrograph of a completed nanowire bioFET device showing characteristic trapezoidal cross section due to anisotropic etch [23].

Last steps in the fabrication process include metallisation and photoresist passivation which protects the metal layer from exposure to electrolyte.

3. Whole blood detection

Since their introduction in 2001, electronic-label free sensors have demonstrated a range of qualitative applications ranging from proteins [3], oligonucleotides [18], cellular response [19], viruses [20] and cancer biomarkers [15]. Despite the huge success in the academic community, these sensors have suffered a fundamental limitation in application due to their inability to operate and sense in physiological solution and, more importantly, whole blood. Debye screening, bio-fouling and non-specific binding are some of the reasons, the nanowire bioFET technology has not reached its full potential.

The issue of the whole blood detection can be solved by introducing an in-line silicon-based microfluidic purification chip (MPC) that can be integrated with an existing bioFETs [15]. The chip is pre-functionalised with receptors (antibodies) that can capture specific analytes (antigens) from blood [21]. Using a Bosch etch process [22], the MPC chip is fabricated from bulk silicon wafer and consists of a honeycomb lattice of pillars, 100 μm in diameter and 100 μm high. These pillars serve to enhance the binding surface and therefore increase the capture efficiency of the MPC [Figure 2(a)]. To confer amine functionality to the MPC surface (as well as bioFETs), a 3-aminopropyltrietoxy silane (APTES) is used to modify the surface. Following the amine functionalisation, capture antibodies are immobilised on the MPC surface via photo-cleavable cross-linkers. In the second step [Figure 2(b)], whole blood is injected into the chamber which allows binding of antigens to the corresponding antibodies. After incubation, the chamber is washed and substituted with a low ionic strength buffer and exposed to ultraviolet (UV) radiation. The UV exposure step releases immobilised antibody-antigen complexes [Figure 2(c)]. The sample is then transferred to a nanowire/nanoribbon bioFET chip which is functionalised with secondary capture antibodies which bind different epitope of antigen [Figure 2(d)].

The efficiency of the method as determined by ELISA was found to be around 10%. However, one should bear in mind that this efficiency is obtained from a mere fingerprick amount of blood (10 μl) and very short incubation time (< 20 min). The efficiency can be further improved by increasing the sample size, incubation time and reflowing the sample.

4. Calibration

Over the past years, significant efforts have been made to demonstrate qualitative applications of bioFETs such as detection of biomarkers, oligonucleotides, cellular response and pathogens. However, none of the past work has discussed issues such as reproducibility, repeatability and quantification of analytes. Moreover, the links between bioFET response and its electrical characteristics were not explored.

Two recent papers have already suggested application of solution gating and solution gate transconductance as means to calibrate devices [23, 24]. Furthermore, it can be shown that device response is proportional to the change of the surface potential where the proportionality constant is the device transconductance [25]. Starting from a MOSFET equation in the linear region:(1) under an assumption of , the bioFET current change due to the charge binding is(2) Normalised sensitivity (i.e. sensor response) is therefore(3) while the solution transconductance normalised response is(4) From the last two equations, it follows that current normalised response still depends on device threshold voltage and therefore will be affected by any device-to-device variations. The transconductance normalised response is equal to the change in device threshold voltage which is directly related to the change in device surface potential [1], therefore by performing transconductance normalisation one should expect lower variations in normalised sensitivity for different devices under the same experimental conditions [23]. It is important to notice that previous results are valid only if the total current change (i.e. threshold voltage shifts) leaves the device in the same operation region, i.e. linear. More importantly, if device-to-device variations are small then both current and transconductance normalisation yield the same results [26].

Further expansion of this idea to top-down fabricated devices [26] was done by demonstrating bioFETs with well controlled threshold voltages of approximately 8 mV per die (standard error of the mean, σ_SEM) in the areas where the uniformity of the fabrication processes is the highest, i.e. wafer centre (Figure 3). Using the advantage of the top-down processing techniques over bottom-up techniques in terms of device uniformity, a novel approach of detection was introduced by measuring initial kinetic (device current) rates rather than the end-point detection which is normally utilised in electronic label-free detection schemes.

Figure 2. Microfluidic purification chip (MPC) consisting of honeycomb structure of silicon pillars used by Stern et al. for capture-release filtration of whole blood. (a) Primary antibodies are immobilised on the MPC via photocleavable crosslinker. (b) Whole blood is added to the MPC allowing the binding of specific antigens to the MPC surface. Washing step includes substitution of the whole blood with a buffer solution with low-ionic concentration. (c) UV-exposure allows release of immobilised antibody-antigen complexes into the buffer solution. (d) In the last step, the buffer solution with antibody-antigen complexes is transferred to functionalised nanoribbon chip where the detection occurs [26].

Figure 3. Average threshold voltage dependence of nanoribbonbioFET as a function of its position on a 4” wafer [29].

The justification for using initial kinetic rates can be found by examining a simple receptor-ligand binding using Langmuir kinetics [27]. By assuming a reversible reaction , one can calculate time dependence of the relative surface coverage to be(5) and the rate(6) where and are the association and dissociation constants of the receptor-ligand complex. The change of the relative coverage of the sensor surface at is therefore directly proportional to the analyte concentration provided that the sensor has not been previously utilised, i.e. . This fact allows for direct measurement of analyte concentration at the nanosensor surface under assumption that the system is not diffusion limited [28] and the sensor response is directly proportional to the surface charge [29, 30]. Recently, this method has been further confirmed by measuring binding affinity of protein interactions under the assumption of Langmuir kinetics [31].

Figure 4 shows initial current rate dependence of five nanoribbon devices as a function of both device baseline and transconductance. As expected according to Equations (3) and (4), both dependences are linear. Prior work on bottom-up nanowore FET sensors [23] have demonstrated better results using transconductance normalisation, which is expected according to Equation (3) due to the dependence of normalised current change on . On the other hand, transconductance normalisation is equal to threshold voltage shift due to the binding of surface charge which is directly related to the change of surface potential of a bioFET. The equivalency of transconductance and baseline current normalisation is expected for bottom-up nanowires since they exhibit large device-to-device variation in electrical characteristics. For top-down bioFETs, this variation can be as low as 8 mV (< 1%, σ_SEM).

Figure 4. Initial current rate of five devices from the same die recorded simultaneously as a function of both baseline currents and device solution transconductances at and . The relative standard deviations for initial current rates, baseline currents and tranconductances are 0.7%, 0.3% and 0.6%, respectively. Both fits are linear (y = kx) shown on a log–log scale for clarity [29].

Applying the kinetic rate approach to the detection of cancer biomarkers obtained by the capture-release filtration of whole blood one can determine calibration curves which can be used for analyte quantification. Sensing measurements were performed using two different biomarkers – a prostate specific (PSA) and breast cancer (CA15.3) antigens. Measurements were repeated on multiple devices to inspect reproducibility and repeatability of the proposed method. It was found that the calibration curves were linear in the clinically relevant concentration range and that the relative standard error of the mean was below 10% measured on samples obtained from the same stock solution (Figure 5). In addition, a blind measurement tested using a single device on a different measurement setup demonstrated a good agreement with the calibration curve and concentration obtained using conventional ELISA.

Figure 5. Calibration curves for (a) PSA and (b) CA15.3 show linear device response in the clinically relevant range of analytes. Circular data point represents a blind measurement [29].

5. Conclusion

Recent advancements in fabrication techniques and chip integration have opened-up further improvement of the bioFET technology. By integrating silicon based microfluidic capture-release chip for whole blood filtration with silicon-on-insulator bioFET arrays, electronic label-free detection approach has been brought a step closer to point-of-care commercial applications.

Acknowledgements

Authors would like to acknowledge research support of the National Institute of Health under Grant No. 1RO1EB008260-1 and the Defense Threat Reduction Agency under Grant No. HDTRA1-10-1-0037. AV thanks Nitin K. Rajan for critical reading of the paper.