!1

Special Lecture Series and Instrumentation

Lecture 8 - FET Based This lecture looks at the Ion Sensitive Field Effect Transistor (ISFET) which is the basic chemical that can be made using the same basic structure as a MOSFET, but with some important changes. The initial implementation of the ISFET was a pH sensor operating in a potentiometric mode and that’s what this lecture presents first. We will then go on to look at different instrumentation methods for the ISFET, some of which are suitable for integration with CMOS processes for single chip microsystems or the creation of sensor arrays. Issues with ISFET measurement will be introduced along with techniques to deal with these with differential instrumentation. In the subsequent lecture we will look at some alternative implementations, for example using weak inversion mode of operation, additional sensing modes, or nano-wire structures. An example of a sensor chip developed at Edinburgh which combines pH, pO2 and amperometric biosensing capabilities will also be covered.

Ion Sensitive Electrodes We begin by reviewing the concept of ion selective or ion sensitive electrodes, potentiometric devices where the electrode potential is measured when there’s no current flowing. They are characterised by the the Nernst equation: RT a E = E0 + ln i1 nF a ✓ i2 ◆ This depends on the ratio of activities (approximated as concentrations in dilute solutions) of the oxidised (ai1) and reduced species (ai2) being measured. Glass membrane electrodes, which you should be somewhat familiar with are a common form of potentiometric ISE but they’re difficult to miniaturise and so there’s been a lot of activity looking at trying to make a solid state equivalent.

The ISFET The first solid state ISE was developed by Prof. Piet Bergveld at the in the 1970s. The Ion Sensitive Field Effect Transistor (ISFET) is a standard MOSFET with the gate metal/polysilicon removed and replaced by the solution being measured and an electrochemical reference electrode. The gate dielectric, commonly SiO2, acts as the ion sensitive membrane where the surface potential is controlled by the pH of the solution. We will begin by revising some of the formulae for the operation of a standard MOSFET, that also apply for the ISFET. The drain-source current of the device when it’s operating in the linear region is defined in terms of the gate source voltage VGS, the threshold VT, the drain source voltage VDS and a device dependent gain β which is made up of terms representing carrier mobility μ, the gate oxide capacitance Cox and the dimensions of the channel region. V 2 I = (V V )V DS W D GS T DS 2 = µCox  L With the potential difference between the source and the substrate/bulk set to 0V, the threshold voltage is made up of the flatband voltage, a term representing the gate voltage due to the charge in the depletion region QB and the potential due to the doping of the bulk ϕF. !2

QB VT = VFB + +2F Cox Finally, the flatband voltage is made up of terms representing the work function difference between the “metal” gate and the silicon substrate and the potential due to fixed charges at the silicon/oxide interface (Qf or sometimes Qss) and charge incorporated into the oxide Qox.

Qf + Qox VFB = MS Cox In the ISFET the effective gate source voltage is controlled by the reference electrode connection to the solution but the characteristics of the solution will also affect the threshold voltage. The expression for the VFB of the ISFET is:

Qf + Qox VFB = Eref + ⇥0 + sol Si Cox The first three terms here are the contribution from the “gate” side of the device, equivalent to ΦM in a MOSFET. This includes figures for the potential of the reference electrode (Eref) and the solution dipole potential (χsol), but the most important for operation as an ion sensor is the surface potential Ψ0. The final two terms are due to the work function of the silicon and the oxide charges as before. The interface between the thin gate oxide over the channel region of the transistor and the solution will consist of hydroxyl (OH) groups, which can either accept or donate protons (H+ ions) from the solution. They are “amphoteric” sites meaning they can exist in acidic, basic and neutral forms. The balance of these charges at the surface will depend on the pH in the solution and will effectively change the surface potential of the oxide. The equilibrium reactions between the surface and the solution are as follows: + SiOH SiO +H , B SiOH+ SiOH + H+ 2 , B The surface acts as a buffer for changes in the pH of the bulk solution. When this increases the surface will donate protons, becoming more negatively charged whereas when the pH reduces the surface will accept protons and become more positively charged. The theory is developed in more depth than is appropriate for this course in [1] and you are directed there if you’re interested in finding out more. Basically though the surface potential at the oxide solution interface (Ψ0) is dependent on the pH of the bulk solution (pHB) with the following function: ⇥ kT 0 = 2.3 ⇥pHB q This is Nernstian in nature but includes a sensitivity parameter α which can vary between 0 and 1. The closer to 1 it is then the more Nernstian the pH response, i.e. it’s closer to acting like a glass membrane ISE,where ∆Ψ0 = -59.2 mV/pH at 298K. Theoretically the sensitivity α will have the following formula: 1 ↵ = 2.3kT Cdl 2 +1 q int This is a function of physical constants (k, T & q) as well as the variables Cdl and βint which represent firstly the double layer capacitance at the solution/oxide interface and secondly the buffer capacity of the oxide surface. !3

The capacitance Cdl is a result of the electrical double-layer that forms at the surface of any object when it is submerged in a liquid. In the most common model of this, known as the Gouy-Chapman-Stern model, this is made up from two parallel layers of charges. For our ISFET, the inner part is a tightly bound layer of ions at the oxide surface referred to as the compact Stern or Helmholtz layer. This is surrounded by a diffuse layer of charges that balance the surface potential or charge of an electrode or other charged surface when in contact with a solution. These are effectively two capacitances in series, one with a fixed value representing the Stern/Helmholtz layer (CH) and one representing the diffuse layer (Cd) which will vary with the solution concentration. The characteristic thickness of the double- layer is defined by the Debye length which is inversely proportional to the square root of the concentration. Therefore, the overall DL reduces in thickness as the concentration increases and, as it’s considered to have a constant permeability, the capacitance of the layer will also increase. So Cdl will vary with changes in the ionic concentration of a solution that are unrelated to pH, which will change α and affect the ISFET performance.

The intrinsic buffer capacity of the oxide surface (βint) is a measure of the ability of the surface to accept or donate protons to and from the solution. It should be maximised to increase the sensitivity by making α closer to unity. Silicon dioxide isn’t the best material to use for an ISFET as the buffer capacity is relatively low. This means the sensitivity is sub- Nernstian and the ISFET made with SiO2 is more sensitive to changes in Cdl. Other materials like silicon nitride (Si3N4), aluminium oxide (Al2O3) and tantalum pentoxide (Ta2O5) are better for a variety of reasons and have higher βint and α. One reason for the better performance of the aluminium and tantalum oxides might be the greater number of oxygen atoms involved in the oxides, meaning a greater density of amphoteric sites at the surface. The graph on slide 12 is taken from [2] and shows the response curves of made with the different materials mentioned previously. It shows how increasing the buffer capacity through choice of the gate dielectric can make the ISFET response more linear and more Nernstian. The best result is obtained for tantalum pentoxide which has a sensitivity that is almost Nernstian and appears to be very linear. The graph on slide 13, taken from the same paper, shows the affect of changing the ionic concentration of sodium chloride in the solution while keeping the pH constant. Silicon dioxide shows the worst cross correlation while Ta2O5 is much more insensitive, as the very large buffer capacity prevents Cdl affecting the ISFET response.

As we’ve seen, the oxide/solution surface potential (Ψ0) is dependent on pH and therefore so are VFB and VT. If all else is constant the threshold voltage will have the same dependence on pH as the surface potential: ⇥V kT T = 2.3 ⇥pHB q This means that the ISFET can be completely controlled by the solution pH, though setting it up for a controlled measurement can be quite complicated as we’ll see in the next section.

ISFET Instrumentation A standard method for measuring the ISFET is to bias it in the linear region of operation, so VDS is set to be some small, constant value so that the transistor channel isn’t pinched. Then the current through the device, IDS, is also controlled so that it is kept constant. If we remember the equation for the current in a MOSFET in the linear mode given earlier in this document, it can be rearranged to give an equation for the gate-source voltage, VGS. The result is that if VT increases due to a change in the pH of the solution, then VGS will also have !4

to increase to balance this and keep VDS and IDS constant. This all requires some sort of electronic feedback to achieve the correct behaviour. The source and drain follower circuit detailed in slides 16-18 is described by Bergveld in [2], here the ISFET channel forms part of the feedback and biasing for an operational amplifier. The input voltage Vin is a constant, so the non-inverting input of the op-amp is biased by R1 and R2 acting as a voltage divider. Following the op-amp rules then V– will be equal to V+ and this is effectively the source-drain voltage (VDS) of the ISFET. Similarly the voltage across R3 will be the same as across R1. As this is a constant then there will be a constant source drain current (IDS) through the ISFET. If the pH changes in the solution that will be reflected as a change in the threshold voltage VT. If VT increases then the only way for VDS and IDS to stay constant will be for VGS to also increase. VG is effectively the reference electrode potential, which is constant, and so the source potential (vs. ground) will decrease proportionately. This is effectively the output voltage Vout. It is possible that if the wrong voltage occurs across the passivation layer that the oxide can be damaged through electrolytic breakdown. The channel region of the device should be kept positive with respect to the solution, for example by controlling the range of source voltages possible. The ISFET should be biased in a “normally-on” condition for example. The diodes in this circuit serve to control the maximum range of outputs to prevent damaging the device. If they are LEDs as shown here they will light up when the system is out of range. The ISFET amplifier is a method of achieving ISFET biasing which was also developed by the originator of the technology, Piet Bergveld [3]. Basically the circuit is an instrumentation amplifier where the ISFET replaces the resistor that sets the gain of the circuit. The input to the instrumentation amplifier is set by the resistor RDS and the constant current Iin which also controls the voltage across the ISFET so that VDS = IinRDS. If VT drops due to an increase in pH then the effective resistance of the ISFET channel will decrease and the output of the instrumentation amplifier, which is inversely proportional to this, will increase. If this is now larger than Vref the output of the op-amp will increase feeding a current through RS that will raise the source voltage, reducing the effective gate voltage and raising the effective channel resistance. This helps keep the source drain current of the ISFET constant. So, the change in the source voltage will be the same as the change in VT but with an opposite sign while the output will be this value amplified by the ratio of Rout and RS. The output of the circuit will be:

Rout Rout Vout = If Rout = VS = VT RS RS Setting up this system is relatively easy. Place the sensor in a buffer with a known pH (typically pH of 7) and adjust the value of Vref to zero the output voltage. This effectively sets the current through the ISFET which is required to make the instrumentation amplifier equal to Vref. Then the sensitivity of the output is set with Rout and this calibration could be performed with another known buffer solution. For example, place the ISFET in a solution with a pH of 4 and adjust Rout to give Vout = –3V for 1V/pH sensitivity.

CMOS ISFET Fabrication There are a number of issues to be considered when deciding how to fabricate ISFET sensors. Standard CMOS usually uses a self aligned process where the gate electrode (poly or metal) is fabricated before the source and drain regions. These are then produced by implantation of dopants to form the source and drain, with the gate electrode and field isolation, acting as a mask. This makes integrating an ISFET in this process a problem as you can’t expect to remove the gate electrode afterwards without damaging the thin gate oxide and you can’t !5 make a FET in this process without a gate electrode. Either you need a custom process that uses some other method to define the S/D or you need a different way to make ISFETs. For example, the SMC used to run a metal gate PMOS process which wasn’t self aligned. This would potentially allow the fabrication of a bare gate ISFET if the metal gate was removed or if the gate area of the sensor was protected from deposition of the metal layers and subsequent dielectric layers if a lift off process was used. This process also used a combination gate dielectric of silicon dioxide and silicon nitride which could potentially produce a better ISFET than a process with a simple oxide gate dielectric. An alternative to this that has been developed to make CMOS compatible integrated ISFETs is to use the metallisation layers available in the process to effectively bring the gate connection up to the surface of the integrated circuit. There the top passivation layer of silicon nitride or silicon oxy-nitride is ideal to act as a pH sensitive layer without having to alter the process. The figure in slide 22 is a cross section through a CMOS compatible ISFET taken from [4] while the original concept for this type of structure was first published in [5]. One issue with CMOS integrated ISFETs is the problem of the body effect for n-channel ISFETs if the process has a common, p-substrate, for all NMOS devices. Typically CMOS circuits will ground the substrate (bulk connection) and the body effect is not a problem because the source contact of the NMOS device will also be grounded in most circuits. The PMOS devices will be fabricated in n-well implant regions and can have independent bulk contacts. Although many modern digital CMOS processes will use multiple wells or other other architectures where every device can exist in it’s own well this is often not the case in older analogue processes used for sensors.

The REFET Correct operation of an ISFET requires a good reference electrode and these are hard to produce as a miniaturised integrated device, unfortunately there is no solid-state reference electrode. There are integrated quasi-reference electrodes which can operate in a limited range of conditions and one of the most common for biosensing applications is the Ag/AgCl electrode. The REFET is not a FET based reference electrode but a possible way to do without one in ISFET sensing applications. The REFET is a device identical to the ISFET to be used in the sensor application but with some additional layer over the pH sensing area which blocks the transport of protons to and from the surface. This will potentially allow the use of a pseudo-reference electrode which doesn’t have a known, fixed potential. A possible example of this is a platinum electrode. The output of a source drain follower using the ISFET will be a function of the pH and the pseudo-reference electrode potential VPRE while the output of an identical circuit using a perfect REFET will simply be a function of VPRE. By feeding these into a differential amplifier we will end up with an output that is solely a function of the solution pH. Unfortunately this is harder than it seems and there have been many problems in producing an ideal REFET where the application of the blocking layer doesn’t affect the matching between the two sensors. It will typically change the threshold voltage of the REFET and can also cause drift if the blocking layer can become contaminated by the solution being measured. Overall this can be a significant barrier to the development of FET based chemical sensors.

ISFET Problems There are also problems, or rather issues to be taken into consideration, with ISFETs in general. They are obviously temperature dependent through the inclusion of the thermal !6 voltage in many terms that feed into the response so adding a temperature sensor is essential. Drift is also a huge problem with silicon nitride sensing layers tending to change to include more oxygen over time when they are in aqueous solutions. The usual solution for this is to measure the drift and correct for it over time which isn’t very elegant and controllable but seems to be an accepted method. Another general problem with electronic sensors in aqueous solutions is making sure of good sealing and encapsulation of electrical contacts to prevent corrosion and other impacts on the stability and operation of the device.

Weak Inversion ISFET Operation The weak inversion or sub-threshold region of operation in a MOSFET or ISFET is increasingly being exploited due to the increased sensitivity it promises. In the weak inversion region the current between source and drain becomes exponentially dependent on the gate voltage, although the overall level of current is orders of magnitude lower. This arises from the coupling of the gate voltage into creating an effectively constant surface potential in the channel region and the fact that (assuming source and bulk are grounded in an n-channel device) this potential controls the forward bias of the source-channel pn-junction. That in turn controls the concentration of electrons in the source end of the channel and therefore the diffusion current between drain and source. The drain current in weak inversion is:

q(VGS VT ) q ID = ID0 exp 1 exp VDS nkT kT ✓ ◆ h ⇣ ⌘i Where ID0 is the current when VGS = VT. This is slightly different to the expression seen previously as it includes a dependence on the drain source voltage. However, at significant values of VDS this term is negligible and can be ignored. In an ISFET the threshold voltage becomes dependent on pH and so we can rewrite this as q kT I = I exp (V 2.3↵ pH) D D0 nkT GS q  Where γ includes the constant parts of the equation for VT in an ISFET, such as the reference electrode potential. Now we have an exponential dependence of current on pH, which should be remembered is also an exponential function of the hydrogen ion concentration. This could potentially be exploited using translinear circuits (beyond the scope of this course) to obtain outputs that are linear functions of the hydrogen ion concentration as detailed in [6]. In a CMOS integrated ISFET the effective sensitivity will be dependent on the slope factor n, which in-turn is dependent not only on the capacitive potential divider between the gate oxide and the depletion layer capacitance, but also the other capacitances in the ISFET structure. This includes the sensing passivation layer at the top of the floating gate metal stack and the electrical double layer capacitance in the solution. This effectively lowers the drain current sensitivity to changes in pH but could be thought of as a positive effect which increases the effective range of gate bias over which weak inversion operation occurs. This is explored more deeply than there is space for here in [7], from where the CMOS ISFET results in slide 30 are also taken.

Other FET Based Sensors The ISFET can be made sensitive to other ions by changing the gate dielectric to a different material, or by applying it as a coating to a standard pH ISFET. It is also possible to make an !7

FET sensor for non-ionic chemistries with a sensing layer that changes the local pH or changes the charge distribution at the gate surface so that it can be sensed with a FET. Enzyme based FET sensors will use a similar operating principle and may use the same type of enzymes used to make conductivity sensors, such as urease. The enzymatic reaction must produce protons for sensing with a pH sensitive ISFET. A REFET type differential measurement can be obtained by combining a standard ISFET with with the ENFET. There are issues with the sensitivity being very non-linear making quantitative measurements difficult and there are methods using electrochemical feedback to adjust the local pH in the ENFET to make it more controllable. For more information see this review paper [2]. Other Bio-FET sensors can use biological sensing methods running from antibody attachment in an “immuno-FET, DNA binding or right up in scale to whole cell sensing. Binding of charged molecules like DNA is particularly suitable for FET based sensing. A review of bio- FET sensors can be found in [8].

METOXIA Chip Case Study and the O2-FET The sensor chip shown as a CAD layout on Slide 34 was designed at Edinburgh as part of the EU FP7 project “METOXIA”. It includes ISFET based sensors for pH and pO2 as well as a set of three potentiostats for amperometric electrochemistry. The intention was to use it to sense the chemical environment in a cell culture environment to investigate cancer/tumour cells. This is considered important because many types of tumour maintain a low oxygen, high pH microenvironment which can increase their resistance to radio and chemo-therapy. The ISFET structure is a standard CMOS compatible setup as we’ve seen before. There are additional electrodes at the surface in the cross section shown in slide 35 but those are only used on the oxygen sensors and we’ll return to that later. The schematic for the ISFET instrumentation is shown in slide 36 which is based on a differential pair of CIMP (complementary ISFET-MOSFET pairs), which were intended to provide a linear amplified measurement of changes in pH without the use of a true reference electrode [9],[10]. The oxygen sensing system on the Metoxia chip is based partly on the operation of the amperometric oxygen sensor known as the Clark electrode. The Clark electrode (which you may or may not be familiar with from other courses) relies on the reactions that occur at the surface of a noble metal working electrode which is held at a cathodic potential in an aqueous solution:

O2 + 2H2O + 2e– → H2O2 + 2OH–

H2O2 + 2e– → 2OH– O2 + 2H2O + 4e– → 4OH–

O2 + 4H+ + 4e– → 2H2O The first two reactions are are two electron reactions, while the third is a four electron reaction but both have the same overall outcome, the production of four hydroxyl ions. These are more likely to occur in neutral or alkaline solutions while the final one is more likely to happen in an acidic solution where there are more protons available. Regardless of this these will all have the same result in that they will raise the pH locally to the electrode where the reactions occur. This can then be exploited to turn a pH sensor like an ISFET into an oxygen sensor by fabricating oxygen transducing electrodes close to the pH sensing gate area. This was originally proposed by Sohn and Kim [11] and later refined by Lehmann et al., [12] who called it an “O2-FET”. !8

In the Lehmann implementation two separate reference electrodes are used, one for the oxygen transducer and one for the ISFET. The noble metal electrode (palladium or platinum) is held at a cathodic voltage between –0.65 V and –0.85 V versus the first reference electrode. The ISFET is biased as normal with a constant VDS and IDS and the pH is sensed by measuring the source voltage VS versus the second reference electrode. By modulating the noble metal electrode voltage it is possible to measure both the solution pH and changes in the dissolved oxygen concentration. The Metoxia chip uses a more complicated setup with complementary ISFET-MOSFET pairs arranged into two Wheatstone bridge circuits then measured differentially. This should provide better temperature stability for the measurement though I’m afraid I don’t have any further details of how this instrumentation operates. There’s probably a whole lecture in there! The oxygen sensor relies on the ability to post-process the CMOS integrated circuits to add an additional noble metal electrode which will act as an oxygen transducer. These are shown in grey on the zoomed in layout in slide 41. The platinum electrodes may be deposited as an additive process using focussed ion-beam (FIB) deposition. This avoids the need to perform challenging photolithography on the tiny (3x3mm) METOXIA dice in order to produce patterned metal electrodes.

Silicon Nano-Wire Biosensors I’ve added a discussion of SiNW sensors to this part of the course because the mode of operation of these devices is essentially the same as that of the FET based sensors. Although we’re concentrating on silicon nanowires there are other popular options such as metal oxides (In2O3, ZnO) and carbon nanotubes which operate in a similar fashion. Silicon nanowire devices also resemble advanced 3D MOSFET devices such as the FinFET that have been proposed as a replacement for CMOS in digital electronics in order to continue the Moore’s Law progression. In the FinFET or the nanowire FET the gate electrode completely surrounds a three-dimensional nanoscale channel allowing complete control of the channel conduction without the negative effects seen in modern with deep sub-nanometre gate lengths. In a SiNW sensor the length of the nanowire semiconducting channel is obviously a lot longer than in a FinFET but basic principle is the same. The sensing process involves changing the charge distribution surrounding the nanowire which will modulate the conductivity between the “drain” and “source” electrodes. The nanometre width scale of the channel means that charges on the surface can affect the conductivity throughout the wire leading to much greater sensitivity. As shown in slide 43 in the µm scale regime, the charge on the surface can only affect the conductivity of a small part of the semiconducting channel while in the nm regime the whole channel can be influenced. This increased surface to volume ratio is one of the main attractions for the use of nanowires in sensing. Another key selling point is the possibilities for label free detection represented by SiNW biosensors. This does however depend on the bio-recognition event leading to a change in the charge distribution at the nanowire surface. For example, adding positive charge to the surface of a p-type semiconducting silicon nanowire will deplete the majority charge carriers and reduce the conductivity while increasing the negative charge will enhance the hole concentration and increase the conductivity. Although the label free detection requires that the target analyte change the charge distribution this can happen in a number of different ways. Even if the target has no overall charge, it may be that the portion closest to the surface of the wire when it is bound to the receptor does have a significant charge and can affect the nanowire conduction. If you’re interested in a review article on the subject of SiNW biosensors I would recommend [13]. !9

SiNW Fabrication There are two approaches to the fabrication of nanowires, which you may be familiar with from the lab-on-chip technologies course. Bottom up fabrication of nanowires may be a little different to biologically inspired styles from the LoC course but it does involve the “growth” of the wires. The standard way to do this is to use a “Vapour-Liquid-Solid” CVD process which starts with a silicon substrate and the deposition, either randomly or through nano- lithography, of catalytic nanoparticles which are often made of gold. The substrate is exposed to a gas containing a compound of silicon at a high temperature where the gold nanoparticles are liquid and can form a eutectic alloy of silicon and gold. This will continue to absorb silicon from the vapour phase until it becomes supersaturated at which point silicon will precipitate as a crystalline solid at the interface with the substrate. This forms a whisker which grows up from the surface. The doping of the nanowire can be controlled by the introduction of appropriate contaminants during growth and the length can be controlled by the process parameters. Once the nanowires have been grown they can be oxidised in-situ to form a gate dielectric. This is a relatively well controlled mature process but the question then becomes one of how to use these nanowires to build a sensor. As with CNTs it may be relatively easy to make them but turning them into a transistor or a sensor is much more challenging and complex. The first step is obviously to cleave them from the silicon substrate then to re-deposit them only another surface. This is often another silicon wafer coated with an oxide layer which can act as a back-gate dielectric with the Si substrate as the gate electrode. There are many different methods proposed to allow the alignment of the nanowires on the surface which we’ll not go into here but often they still produce a relatively random distribution. In order to turn the wire into a sensor we need source drain electrodes and the figure in slide 45 shows this being done schematically using a lift off process to produce gold electrodes. As you might imagine this is quite a bespoke process and it is difficult to imagine how regular arrays of these devices could be produced. However, this is an area of considerable research and there may be new techniques in the future which can exploit the advantages of bottom up nanowire fabrication. The main advantage is obviously that it does not require specialised nanoscale lithography to produce. This is not the case with the standard vision for a top-down fabricated nanowire which requires advanced nanofabrication techniques. A typical implementation would begin with a silicon on insulator substrate with device layer of the desired thickness for the nanowire being fabricated. The first step is to dope this with the required nanowire dopant level as shown in the process flow on slide 46. A second doping step is then used to create lower resistance source and drain regions before standard photolithography and silicon etching is used to define the S/D electrodes and a relatively crude channel region. The actual nanowire is formed in the next step which requires either advanced photolithography or another technique such as electron beam lithography. The final steps would typically involve the growth of a thin gate oxide over the nanowire (which also serves to reduce the wire cross section) and the addition of metal contacts to the source and drain. The requirement for nano lithography has limited the application of top-down nanowires because of the expense and the fact that few sites have this capability. This has fuelled the development of ingenious techniques to fabricate nanoscale silicon structures without the need for nano lithography. We will look at two possible techniques here. The first of these uses something similar to the process described above but requires the use of an SOI substrate with a {100} surface and careful alignment of the initial pattern to the {110} crystal planes. Silicon nitride is used as a mask for lattice or plane dependent etching using a wet solution of an alkaline silicon etchant such as TMAH (tetra-methyl ammonium hydroxide). This etches faster in the [100] and [110] !10 directions than it does in the [111] direction and will tend to leave exposed {111} surfaces. The angle between the [111] and [100] directions is ~54.7° and so the etched structure will have a trapezoidal cross section. The exposed {111} surfaces are oxidised to passivate them before the silicon nitride is removed from the nanowire region but not the source and drain contact areas. A second PDE is then used to etch away more of the silicon to leave a pair of nanowires with a triangular cross section. The size of these can be tuned by overetching against the oxide protected outer surface before the thick oxidation is removed by another wet etch and a final gate oxide sensing layer is grown. The final step is to remove the silicon nitride protection from the S/D regions and deposit metal contacts. This process can produce single crystal silicon nanowires without advanced lithography but the use of wet etching of the device layer means that it is not generally considered to be CMOS compatible. An alternative to this which may be considered to be more compatible with CMOS processing is based on the techniques used to create silicon nitride spacers on either side of the gate electrode in a modern CMOS process. In this process a “dummy gate” of a dielectric material is formed on a passivated silicon substrate before a low-pressure chemical vapour deposition (LPCVD) process is used to deposit a conformal coating of amorphous or polycrystalline silicon. After appropriate ion implantation to determine the doping in the silicon, anisotropic reactive ion etching is used to thin the polysilicon layer until it is removed from the top of the dummy gate and from the majority of the substrate. However, due to the conformal coating process this will tend to leave nano-wires of silicon along the edges of the dummy gate mesa. The initial versions of this process tended to produce wires with a rounded cross section, as is common in the SiN spacer process but subsequently the nanotechnology centre at the University of Southampton have developed a process using amorphous silicon (which is annealed to form polysilicon) and a highly anisotropic etch process to produce wires with an extremely rectangular cross section as shown in the SEM image in the lecture [14]. These SiNW sensors could potentially be produced as part of a CMOS process, though probably not without design rule changes, but it is most likely that they could be integrated with CMOS electronics through post processing. We will return to the Southampton nanowires to look at some results of biosensors later.

Biosensing with SiNW The requirements for a biosensing element which will work with a silicon nano-wire sensor are similar to those described earlier for the Bio-FET sensors. However, the high sensitivity and nanometre scale interaction lengths involved in NW sensors have the potential to enable a wider range of label free technologies. Enzymatic sensors, for example for glucose, would probably operate by changing the pH locally to the nanowire, in a similar fashion to an EnFET. One possibility for making a better SiNW pH sensor is to graft (3- Aminopropyl)triethoxysilane (APTES) molecules to the -OH groups on the SiO2 surface in a process called silanization. The APTES molecule has an amine (NH2) group which can act as a site to buffer changes in the pH of the solution but will provide a more linear potential dependence than silicon dioxide. This is also ideal as a linking molecule for the attachment of biorecognition elements to form other types of . Other biosensors using SiNW will require that the molecular recognition event changes the charge distribution at the nanowire surface leading to a change in the conductivity in the channel. Antibody antigen interaction is one example of significant interest in the production of sensitive sensors for biomarkers of disease. One example is prostate serum albumen (PSA) which is a biomarker for prostate cancer. DNA and RNA recognition is relatively straightforwards due to the negative charge that single strands of these molecules exhibit. !11

Slide 51 shows an example of an enzyme based biosensor using a silicon nano ribbon [15]. The reference to this being a nano-ribbon rather than a nanowire demonstrates that it was fabricated using relatively low cost microfabrication techniques such that the width is in microns rather than nanometres. The thickness of the conducting ribbon is on the nano-scale. Slide 52 shows the results of sweeping the back gate voltage on the Southampton polysilicon- NW sensor, at various values of drain-source voltage. This is a p-doped nanowire so setting the back gate (substrate) to a negative potential compared to the source gives an increase in conductivity. The sub-threshold characteristic of this nanowire shows a swing of 2-3 V/ decade, depending on the value of VDS, which reflects the relatively thick back gate dielectric layer. The next slide shows the operation of the Southampton NW as a biosensor. The oxidised surface of the nanowire is silanized before attachment of an antibody for the inflammatory biomarker TNF-α. During the biosensing experiments the back gate voltage was set to –10 V, with a drain voltage of 5 V and a source voltage of 0V. The graph shows changes in the normalised conductance (G – G0)/G0 (%) during a titration experiment where the concentration of TNF-α is increased from 10fM to 100nM. The numbers 1-7 in the graph represent the time points where the concentration is increased by an order of magnitude. The TNF-α antigen has an overall positive charge under the conditions shown, which should mean that that conductivity reduces as more of it binds to the antibody. However, the silicon nanowire sensor is only sensitive to charges less than a Debye length away from the surface. This represents the screening effect of the ionic concentration of the bulk solution, similar to the effective thickness of the diffuse part of the electrical double layer. This places a limit on the maximum ionic concentration of the medium used in SiNW biosensor measurements, if it’s too high the Debye length will be too short and changes in the charge beyond this distance will not affect the nanowire conductivity. In the case of the Southampton polysilicon NW biosensor experiments the effect of the solution is to effectively screen the positive charge of the antigen leading to an increase in the conductivity as more of the analyte binds to the antibody. Obviously this indicates the difficulty of making assumptions as to the behaviour of these extremely sensitive sensors. The final example here is a DNA sensor using the silicon nanowire sensors developed at the University of Twente [16]. The similar, but uncharged Peptide Nucleic Acid (PNA) can be used to specifically bind to DNA/RNA sequences but is considered a better option to produce sensing elements because the lack of charge repulsion means stronger binding. In the context of a silicon nanowire sensor it also means that the biosensing element does not itself affect the charge distribution. The nanowires are boron doped, and the binding of negatively charged target DNA to the PNA probe will increase conduction. The bulk of the the silicon on insulator substrate forms a back gate which can be used to tune the response of the sensor to give the best sensitivity and linearity. These researchers have also demonstrated a differential mode of measurement using parallel pairs of nanowires with a split source electrode. All measurements are made by applying a small amplitude AC signal between source and drain and measuring the current with a lock in amplifier. The differential measurements are not susceptible to significant drift seen with the standard sensor. However, this paper did not show active sensing using the differential mode. It is unclear how to arrange for one sensor to be sensitive to the desired measurement quantity while the other is insensitive.

References [1] “ISFET, Theory and Practice”; P. Bergveld, IEEE Sensor Conference Tutorial, pp. 1-26, 2003. !12

[2] “Thirty years of ISFETOLOGY...”; P. Bergveld, Sensors and Actuators B, vol. 88, pp. 1-20, 2003. [3] “The operation of an ISFET as an electronic device”; P. Bergveld; Sensors and Actuators; Vol. 1, pp. 17-29, 1981. [4] “Design of a Single-Chip pH Sensor Using a Conventional 0.6µm CMOS Process”; P. Hammond, D. Ali and D. R. S. Cumming; IEEE Sensors Journal, Vol. 4, No. 6, pp. 706-712, 2004. [5] “Ion-sensitive field-effect transistors fabricated in a commercial CMOS technology” by J. Baussels et al., Sensors and Actuators B, vol. 57, pp. 56-62, 1999. [6] L. Shepherd and C. Toumazou, “Weak Inversion ISFETs for ultra-low power biochemical sensing and real-time analysis,” Sensors & Actuators: B. Chemical, vol. 107, no. 1, pp. 468– 473, May 2005. [7] P. Georgiou and C. Toumazou, “ISFET characteristics in CMOS and their application to weak inversion operation,” Sensors & Actuators: B. Chemical, vol. 143, no. 1, pp. 211–217, Dec. 2009. [8] M. Schoning and A. Poghossian, “Bio feds (field-effect devices): State-of-the-art and new directions,” Electroanalysis, vol. 18, pp. 1893–1900, 2006. [9] A. Morgenshtein, L. Sudakov-Boreysha, U. Dinnar, C. Jakobson, and Y. Nemirovsky, “CMOS readout circuitry for ISFET microsystems,” Sensors & Actuators: B. Chemical, vol. 97, no. 1, pp. 122–131, 2004. [10] V. Chodavarapu, A. Titus, and A. Cartwright, “CMOS ISFET microsystem for biomedical applications,” Sensors, 2005 IEEE, pp. 4 pp. EP–, 2005. [11] B. Sohn and C. Kim, “A new pH-ISFET based dissolved oxygen sensor by employing electrolysis of oxygen,” Sensors & Actuators: B. Chemical, vol. 34, no. 1, pp. 435–440, 1996. [12] M. Lehmann, W. Baumann, and M. Brischwein, “Simultaneous measurement of cellular respiration and acidification with a single CMOS ISFET,” Biosensors and Bioelectroncs, vol. 16, pp. 195–203, 2001. [13] K.-I. Chen, B.-R. Li, and Y.-T. Chen, “Silicon nanowire field-effect transistor-based biosensors for biomedical diagnosis and cellular recording investigation,” Nano Today, vol. 6, no. 2, pp. 131–154, Apr. 2011. [14] M. M. A. Hakim, M. Lombardini, K. Sun, F. Giustiniano, P. L. Roach, D. E. Davies, P. H. Howarth, M. R. R. de Planque, H. Morgan, and P. Ashburn, “Thin Film Polycrystalline Silicon Nanowire Biosensors,” Anal Chem, vol. 12, no. 4, pp. 1868–1872, Apr. 2012. [15] Mu, L., Droujinine, I. A., Rajan, N. K., Sawtelle, S. D., & Reed, M. A. (2014). Direct, Rapid, and Label-Free Detection of Enzyme Substrate Interactions in Physiological Buffers Using CMOS-Compatible Nanoribbon Sensors. Nano Letters, 14(9), 5315–5322. doi: 10.1021/nl502366e [16] De, A., van Nieuwkasteele, J., Carlen, E. T., & van den Berg, A. (2013). Integrated label- free silicon nanowire sensor arrays for (bio)chemical analysis. The Analyst, 138(11), 3221– 3229. doi:10.1039/C3AN36586G