Abstract
We will start from the general requirements in chemical and biosensors and then narrow down the specific advantages of CMOS implementation in view of signal transduction from the biological to the electronic domains. As the amperometric sensing on CMOS is similar to most popular mixed-signal designs, we will only focus on field-effect sensors with a polarizable electrode or interface. A general device based on the neuromorphic principles in the previous chapters will be presented for its structure, operation, and circuit models. Variations in device implementation will be examined under the unified neuromorphic circuit model. The interface between the electrode and the buffer media will be modeled with an electrical network. We then present sample measurements in different buffer media and examine the current difficulties in realistic operations with long-term reliability. We will conclude at future challenges and outlook for the biological interface to the CMOS world.
1 Introduction
The vision of Internet of Things (IoT) has forecast thousands to millions of sensors in the local area network to provide ambient intelligence. We will first consider the general requirements in electronic sensors before we introduce what to aim for in CMOS-based biosensors. Electronic sensors convert a specific ambient signal to an electrical signal of voltage or current. Although most of the “signals” in the environment are continuous values represented by floating points, often the final sensor output will be in digital form to facilitate further processing and communication. Therefore, the first stage of a sensor system is often sampling where the signal is collected, held, and reset for the next sampling point, and the last stage is the analog-to-digital (AD) conversion. From the system point of view, we need to define carefully what the target information is at the sensor output, and any other factor that can change the sensor output will be categorized as interference or noise, whether the unwanted signal is systematic or random. Typical noises in sensors include white noises like thermal and shot noises (white noises mean the noise frequency spectrum has nearly uniform distribution), flicker noise that originates from discrete particle impinging at interface, and quantization noise from the AD converter. From this point of view, electronic sensors share many common properties with multi-access radio receivers.
With the signal and noise definition in mind, sensing is NOT merely a “Eureka” event (I find it!), but the sensor performance has clear evaluation metrics:
-
Sensitivity: The minimal quantity of the target analyte that can be detected. This is related to the false-negative statistics, i.e., the target exists but cannot be correctly detected.
-
Specificity: The signal-to-noise ratio (SNR) when the target analyte is compared to noises and interferers. For example, a pH sensor output should be specific to acidity and not influenced by salinity. This is related to the false-positive statistics, i.e., the target does not exist but detection is positive due to other factors.
-
Operational range: The ratio of maximum and minimal target values that can be distinguished. The maximum is often set by sensor saturation and the minimum by sensitivity above. This is similar to the concept of the dynamic range in radio receivers.
-
Bandwidth (Hz): The sampling rate that the target analyte can be reliably collected.
-
Reproducibility: The consistency of the constant or cyclic sensor output over repeated measurements. This will affect the calibration method as well.
-
Long-term stability: As an example in biosensors, the stability can be characterized by the lack of slow drift due to surface fouling or aging as a result of ion contamination and coating material fatigue/plasticity.
-
Power consumption (W or Joule): The power consumed during the continuous sampling or the energy of each sampling.
-
Manufacturing cost: The cost of the sensor system from sampling to final electrical output.
As an example, a biosensor may be deployed to measure the DNA segment concentration from a specific pathogen lysis for diagnostic purposes (Yager, Domingo, & Gerdes, 2008). The full procedure may involve several sequential steps (Allain, 2000):
-
1.
Sample preparation: from patient blood to pathogen lysis,
-
2.
Signal amplification: generation of more copies of targets if necessary, e.g., polymerase chain reaction (PCR),
-
3.
Signal transduction: the sensor sampling the concentration through optical, mass, or electrical methods,
-
4.
Signal readout: conversion to a digitized output.
Here the signal is either the number of the original pathogen cell counts as a diagnostic measure or simply a yes–no answer for infection exposure. The interferer can come from other nonspecific DNA segments, proteins, or molecules of similar molecular weights. The noise can come from salinity variation, readout background, and amplification noise during PCR. Alternatively, we may hope to measure the immunization response to such pathogen invasion directly. The signal is then the antibody biomarker concentration in the patient sample. Although the purpose of diagnostic is similar, immunological sensors have different noise and interferers to their sensitivity and specificity metrics.
In a conventional pathology lab, the four steps earlier involve standard biological and biomedical equipment and protocols such as incubator, autoclave, centrifuge, and microscopes. Not only the equipment is relatively large (though with mass production not very expensive), but also regulation on operator training is stiff. A conventional biosensing procedure needs significant time (such as 1–2 days) if incubation is needed in the sample preparation and signal amplification steps. With the advances in Lab-on-a-Chip (LoC), all four steps can be integrated with the help of microfluidics to finish the entire diagnostic process. However, Steps 1, 2, and 4 are out of the scope of this book, and we will only focus on signal transduction as the sensor function later.
2 Integrated Sensors: Amperometric or Field Effect
To convert chemical and biological signals from the target in electrolytes or biomedia to the electrical domain, three types of electrochemical sensors are commonly adapted (Janata, 2003): Amperometric sensors for the measurement of current with low impedance electrodes, potentiometric sensors for the measurement of voltage with very high impedance and hence nonamperometric (no current flow), and chemiresistive sensors for the measurement of resistance when the target penetrates into the sensing media. Various chemical sensors have been made along the way for over 40 years (Gardner, Varadan, & Awadelkarim, 2001; Kress-Rogers, 1997; Madou & Morrison, 1989) with different approaches, including chemiresistors, chemicapacitors (Steiner et al., 1995), microelectrodes (Kovacs et al., 1994; Lonergan et al., 1996), microcalorimeters, surface-acoustic-wave (SAW) devices (Wenzel & White, 1988), chemomechanical sensors (Lang et al., 1999), and field-effect-transistor (FET)-based sensors (Colapicchioni, Barbaro, Porcelli, & Giannini, 1991; van der Spiegel, Lauks, Chan, & Babic, 1983). Some of the sensors such as chemiresistors and SAW-based sensors are difficult to integrate with CMOS, while some sensors such as microelectrode arrays and FET-based sensors are ready for CMOS integration.
Chemical and biological sensors integrated with CMOS transistors have fundamental advantages in operation and cost. Existing CMOS technology provides ready analog-to-digital converter for sampling, fast computational speed for signal processing, large data-storage capacity, accurate control circuitry, standard wired (e.g., USB) and wireless (e.g., Bluetooth) communication support, and high reliability on the affordable mass production. The sensing electrodes, amperometric or nonamperometric, can be very small if needed by the sensing application, although most of the time the choice of sensing gate area is governed by the noise consideration instead of what can be achieved in production. Many of these factors can be used to improve the sensor performance defined earlier.
For amperometric microelectrode arrays (MEA), the sensing principle is to measure an analyte-related current, either a natural current such as the ionic current in the neuron action potential cycles or an induced current by applying a test voltage such as the cyclic voltammetry. To complete a DC current path in the buffer fluid, the electrode has to induce a redox reaction at the surface for the continuity of electronic and ionic currents. This can be the target analyte such as oxidation of glucose or some balance provided by the electrode such as the Ag/AgCl reference electrode.
Note that buffer ion impinging will cause AC noises or transient IV characteristics of field-effect sensors, but not DC current. When the surface potential is higher than 1.2 V, water electrolysis and bubble formation are possible hazards in saline when a non-negligible DC current exists. This makes field-effect sensing attractive, especially for in vivo scenarios. The sensitivity of the MEA (Berdondini et al., 2005, 2006) is similar to standard measurements of electrical currents in a large DC range, e.g., from pA to mA by source-measuring units (SMU), or in a large AC range, e.g, from 0 to −90 dBm by radio receivers. Isolation of the leakage current path is similar to the practice in digital or analog circuit design, as the frequency of interest is often much lower than 10 MHz and all devices are in quasi-static modes. Therefore, we will limit our attention in this book to FET-based chemical and biosensors.
2.1 Floating-Gate-Based Devices as Sensors
The ISFET (ion-sensitive field-effect transistor), introduced in 1970 by Piet Bergveld is the first example of CMOS-based chemical and molecular sensors (Bergveld, 1970). An FET with a gate open to solution for signal transduction and the drain current as the readout is shown in Fig. 12.1a. “Ion sensitive” indicates that charged species in solution contribute to the operation of the device. The original ISFET has the gate oxide or the poly1 gate directly exposed to the electrolyte, while a possible modification is to have the gate coupling through the interconnect structure, and use the top metal (or another functional coating on the top metal) as the sensing gate. This is shown in Fig. 12.1b. This is often referred as the extended-gate ISFET or floating-gate ISFET. Note that the extended gate will assume one potential in the quasi-static operation determined by all of its nearby structures, but the capacitive coupling to the channel will determine the output current. This will become clear in the neuron MOS equations in the next section.
The concept of adapting the “floating gate” for sensing purpose has been explored further experimentally (Bousse, Shott, & Meindl, 1988; Smith, Huber, & Janata, 1984; van der Spiegel et al., 1983). For example, the extended gate structure in (van der Spiegel et al., 1983) is used as a signal line with one end coated with the chemically sensitive membrane, and the polysilicon layer on top of the floating gate serves as a guard/shield for the signal line. The floating gate in Smith et al. (1984) and Bousse et al. (1988) is covered with Si3N4, which is adapted as the sensing layer in direct contact with the fluid buffer. To unify the operational principle and circuit models in ISFET-based sensors, we will use the relatively complex structure and circuit elements of chemoreceptive neuron MOS (CνMOS) transistors (Shen, Liu, Lee, Minch, & Kan, 2003) as an illustration, and then see the different approximations of sensing gate, control gate, and channel in various possible implementations.
As the field effect sensing is nonamperometric, we will first assume the sensing signal will come from capacitive coupling to the extended floating gate. If the extended gate is directly driven from a sensing potential, we can just use the approximation of infinite sensing capacitance. Furthermore, the total amount of charge at the floating gate can be configured by electron injection mechanism introduced in previous chapters for embedded flash, which can further enhance the signature analyses and the potential fluidic actuation (Jayant, Rajwade, Pollack, & Kan, 2010; Shen, Liu, Jacquot, Minch, & Kan, 2004). CνMOS transistors take the combined advantages of ISFET-based sensors (Bergveld, 1970), chemicapacitors (Steiner et al., 1995), and neuron MOS transistors (νMOS) (Shibata & Ohmi, 1992), and can be integrated into standard CMOS with a simple postprocessing of sensing surface functionalization and fluidic isolation. The extended floating gate in CνMOS transistors inherits the operational advantages of the nonlinear I–V in FET-based sensors: high sensitivity can be achieved through current differentiation and large dynamic range from voltage referencing.
2.2 Generalized CνMOS FET-Based Sensor Structure
Fabrication of CνMOS devices consists of two main steps: the physical design of the foundry-fabricated MOSFETs with the extended floating-gate structure where the sensing gate will be exposed in the contact pad step. The second step is to create specific sensing surface if desirable and the fluidic sample isolation in the postprocessing step. The floating-gate structure operates according to similar neuromorphic principles as discussed in Chap. 8. Figure 12.2a illustrates the CνMOS device structure in a possible two-poly process. Poly2 was chosen here as the sensing and control gates, because in the specific foundry process considered the interpoly oxide gives good capacitive coupling, and the design rules give overall compact device area. The interpoly oxide between the Poly1 floating gate and the Poly2 sensing and control gates has thickness of 58 nm, which allows for a larger range of voltage exploration for various target analytes.
From the equivalent circuits in Fig. 12.2b, we can see that the extended floating gate is one circuit node with a potential V FG determined from the various capacitance input, identical to the coupling ratio consideration in Chap. 8. The potential of a floating-conductor structure will be determined from all of its capacitor coupling, normalized by the total capacitance seen from that floating structure. In the extreme case when one capacitor is larger than the sum of all of the other capacitors, the floating gate will basically follow the potential of that big capacitor, which can be seen as a virtual voltage reference. In Fig. 12.2b we have C CG, C SG1, C SG2 and the transistor channel coupling to the extended floating gate. Also, if we fix the output drain bias (often set at a relatively low value of 50–100 mV to keep the transistor in the linear range and to avoid errors introduced from the drain coupling), the readout current I D is entirely determined by V FG. We can have at least three potentiometric operation modes: (a) fixed V CG and use I D to derive V SG, (b) obtain V CG − I D sweeps to derive V SG, (c) fixed I D and use the compensation V CG to derive V SG. Operation (a) has the simplest setup, but V SG may drive I D into deep subthreshold and therefore difficult to read. Operation (b) is robust in range, but can have larger power consumption during the V CG sweep with unnecessary bulk heating. Operation (c) needs a feedback circuit, but otherwise robust in the differential sensing scheme. Note that we can achieve a capacitive amplification scheme readily when C SG/C CG ≫ 1 in operation (c). If V SG has a known small range as set by the fluidic reference electrode, the control gate can be omitted, and the CνMOS sensor becomes an extended-gate ISFET.
2.3 Structural Variations
For the sensing purposes, as long as the circuits can be represented by Fig. 12.2b, we can have various implementations for CνMOS, and consider other FET-based sensors as equivalent cases of CνMOS. Note that C CG and C SG are defined by the respective coupling area to the floating gates, and they can be implemented by PMOS capacitors as in the embedded flash in Chap. 8 or even Metal 1 to poly, if the coupling area is sufficiently large. The other terminal of the capacitor away from the floating gate can be routed through the interconnect structure to the top metal layer to become the sensing gate, which is similar to the strategy used in Fig. 12.1b. If the sensing gate potential is directly driven by the electrolyte, then there is no penalty in the long wire routing, as long as the sampling frequency is not too high (lower than the inverse of the total RC time constant of that routing line). If the sensing gate potential is affected by the electrolyte through field effects, the routing line is then another “floating gate,” where the potential is also determined by all of its capacitive input or following the largest capacitive coupling.
From the capacitive coupling point of view, the original ISFET in Fig. 12.1a can be seen as the asymptotic cases where C CG = 0 and C SG → ∞ in the CνMOS structure. Another possible implementation can be using the substrate bias as the control gate, as the channel is “capacitively” coupled to the bulk bias in the reverse bias, which is also known as “the body effect.” Many non-CMOS sensor structures such as nanowires can also be modeled by the CνMOS capacitive network as shown in Fig. 12.3a, b, where the target analyte is capacitively coupled to the nanowire channel, and the back gate served as the control gate (Stern et al., 2008). Other possible structural variations in the control and sensing gate placement are shown in Fig. 12.3c–e, which have different design trade-offs. For technology that does not have high-voltage well isolation but with two poly layers, CG and SG can be implemented in poly 2 or metal 1 as in Fig. 12.3c. As the inter-poly and metal-to-poly oxides are relatively thick, the area penalty will be large, and the charge injection voltage will need to be high. The thick oxide can tolerate a large voltage variation, especially from electrostatic charge. As well isolation is not required, many logic processes can directly implement this structure. If high-voltage well isolation or silicon-on-insulator (SOI) is available, then the inverted structures or source–drain–well shorted PMOS in Fig. 12.3d, e can be used.
There is no fundamental difference for the sensing transistor to be in PMOS or NMOS. NMOS is preferred when the source–drain–bulk-shorted PMOS transistor is used as the control gate and sensing gate coupling, as these PMOS capacitors will sit in their own well to achieve the readout transistor isolation. When nonvolatile charge will be tunneled in, the choice of PMOS and NMOS should follow the discussion in Chap. 2 for the preferred injection mode.
2.4 Fluidic and Electrical Isolation
To isolate the electrical connections to the chip and to prevent alkali ion diffusion into the chip, an electrolyte chamber or microfluidic channel is often formed to confine the fluid delivery to the sensing area. For structural integrity, a stiction layer between the chip top surface and the fluidic part is important, especially when pressure is used to deliver the fluidic sample. For prevention of ion contamination, the surface that touches the electrolyte will need to be noble metal such as Au and Pt. Aluminum can also be used when dense alumina has been formed on the surface. Metals have very small permeability for most ions, if corrosion is not a concern. The next choice will be silicon nitride, where the alkali ion permeability is sufficiently low for years of operation. Note that this layer is to prevent ion diffusion, and often a functional layer such as self-assembled monolayer or antibody will be anchored on this layer for specificity.
In the previous two-poly example, stacked vias were opened to expose the poly2 to microfluidic chambers, as shown in Fig. 12.4. The microfluidic chamber is made by silicone elastomer, cured from a master wafer with DRIE patterns, because the elastomer has good stiction property to the spin-on oxide in the IC packaging. The sensing gate is made large to reduce the flicker noise and achieve a capacitive amplification. An alternative example of 2 × 2 sensor pixels by CνMOS is shown in Fig. 12.5 where the pixel area is encapsulated in a fluidic isolation chamber to avoid ion contamination to other parts of the chip.
As an example, the silicone elastomer mixture is composed of 20 parts of the elastomer base with 1 part of curing agent (Sylgard 184, Dow Corning). When the mixture is applied onto the master silicon substrate, a de-airing process is performed in a vacuum chamber for 30 min to get rid of tiny air bubbles during the curing process so as to reduce the surface roughness of the elastomer film, which degrades the stiction. The pattern is transferred from the master silicon substrate to the insulation elastomer layer by further curing the mixture on a hotplate set at 150 °C for 15 min. The encapsulated microfluidic channels for control fluid delivery are then shaped by bonding the patterned elastomer layer on top of the 4.6 mm × 4.7 mm CMOS chip. There are many other ways to make the electrolyte chamber and microfluidic channels, but that is outside the scope of this book.
3 Circuit Models for CνMOS FET-Based Sensors
3.1 Neuron MOS Equations
The neuron MOS equation was first proposed by Shibata and Ohmi (1992) for neuromorphic devices. The characteristic feature of the νMOS transistor is that such a device can be constructed with any number of control-gate electrodes which couple into the floating gate through capacitive coupling (Burns & Powlus, 1996). The floating-gate voltage is established through charge sharing or capacitive voltage division, as a weighted sum of the voltages that are applied to the control gates. The weight on each control gate is directly proportional to the capacitance of that control gate and is normalized by the total capacitance of the floating gate. CνMOS shares the same capacitive network considered in the coupling ratio in Chap. 8, with the sensing gate as one of the inputs.
Capacitive-divider models (Minch, 1997) are adapted from νMOS. As discussed in Chapter 5, e.g. Eq. 5.1, the floating-gate voltage V FG can be expressed as a weighted sum through charge sharing or capacitive voltage division, including those from all control and sensing gates labeled from 1 to X:
where the total capacitance C T can be written as
Q is the total static charge stored on the floating gate. C j and V j are the capacitance and voltage at the jth input gate. The capacitances C ox, C dep, C gs, and C gd follow the same definitions as in conventional MOS transistors. C ox is the gate oxide capacitance, and C dep is the depletion-layer capacitance in the silicon substrate under the gate area. C gs and C gd are the capacitances from the floating gate to the overlapped source and drain region, respectively. C b is the capacitance from the floating gate to the bulk. Q is estimated to be only around pC level to create several volts of the threshold-voltage shift in CνMOS transistors. All potentials are bulk referenced.
Once we know V FG, the output current I D in the subthreshold region can be expressed as
where V th0 is the threshold voltage of the intrinsic transistor. The body coefficient κ is defined as κ = C ox/(C ox + C dep) in the intrinsic transistor and U T is thermal voltage. V S and V D are the voltages at the source and drain. Note that as the floating gate is not directly driven, Eq. 12.3 is the additional correction on the threshold voltage resulting from the coupling between the source–drain and the floating gate. If there is any additional drain-induced barrier lowering (DIBL) by V DS, V th needs to be further corrected.
When V FG is higher than V th, the transistor enters the above-threshold region, and is “turned on.” When V DS = V D − V S < V FG − V th, I D is in the linear region and can be expressed by
Just like the expression in regular MOSFET, there is a discontinuity at V th for I D(V FG) modeled by Eqs. 12.4 and 12.5. This is just because the simplification during the model derivation step. The actual measurement is surely continuous. The discontinuity can be remedied by the EKV (Enz–Krummenacher–Vittoz) model, which is based on a smoothing function that asymptotically approaches exponential and quadratic at the two extremes of subthreshold and above-threshold linear:
The EKV model, although the functional form is complex for intuition, is C − ∞ (continuous for a function and all of its derivatives to the infinite order) and includes the body effect. Therefore, the EKV model is the more appropriate form in circuit simulation. Remember that all biases in EKV are reference to the bulk potential.
For FET-based sensor operations, when we do not need a voltage self-gain in the underlying transistor, we often use the linear or subthreshold region to save power consumption, as excessive self-heating can be problematic to the electrolyte solution. I D is very small in the subthreshold region and can have a limited bandwidth for fast sampling. Therefore, we often use V CG or a feedback circuits to bring the readout to the linear region.
We can define the subthreshold slope S CG seen from the control gate V CG ramp as
The subthreshold slope S CG (in mV) stands for the change of V CG that can cause a decade of change in I D. A smaller S CG corresponds to a steeper slope in the I–V characteristics. In CνMOS devices, C CG represents a constant capacitance depending on the dimension of the control gate, and therefore Eq. 12.7 indicates the subthreshold slope is directly related to C T, assuming the body coefficient κ only has little or no variation during the subthreshold operation.
As expressed in Eq. 12.2, the total capacitance C T includes all capacitance as seen from the extended floating gate. For the CνMOS device with a single sensing gate, the schematic of capacitive loads can be illustrated as in Fig. 12.6. Here of particular interest is the sensing-gate capacitance C SG, which contributes to the change of the total capacitance and hence the subthreshold-slope S CG variation. C SG consists of the capacitance formed by inter-poly oxide and the capacitance by the solid–liquid interaction, i.e., the electrical double-layer (EDL) capacitance C EDL (Adamson, 1982; Bard & Faulkner, 2001). V SG is now the surface potential at the sensing gate. When the fluid has an Ag/AgCl nonpolarizable electrode to set up the fluidic bulk potential at V B, V SG can still be different from V B depending on the ionic profile in the EDL, which can in turn depend on the surface charge, salinity, and steric effects (Kilic, Bazant, & Ajdari, 2007; Liu, Huber, Tabard-Cossa, & Dutton, 2010). The difference between V SG and V B may not be constant either, as the ion profile can change with V SG. If we opt to use a polarizable electrode such as Au, Pt, or carbon fiber, we have to treat additionally the surface potential and ion profile on that polarizable electrode surface. We will look at more details of EDL modeling in the next section.
3.2 The Role of Fluidic Polarizable and Reference Electrodes
To set V FG in the correct operating point, we need knowledge of the fluid bulk potential φ bulk, as shown in Fig. 12.6. Note that the standard Ag/AgCl reference electrode can carry a continuous faradic current (nonpolarizable) and uses the chemical balance between Ag+ and Cl− to set a constant electrochemical potential for Cl−, and for Cl− only. The reference electrode is nonpolarizable where a faradic current can pass with negligible overpotenial and without polarization. This has involved several assumptions (Huang & Huang, 2002; Wong & White, 1989):
-
1.
The buffer media have dominant Cl− as the an-ion.
-
2.
The surface areas of Ag and AgCl nanocrystals are sufficiently large to allow instantaneous (nonrate limiting) redox operation.
-
3.
The overpotential, i.e., the difference between the potential of the half reduction reaction and the potential at which the Ag-to-AgCl redox event is experimentally observed, is negligible.
-
4.
Silver ions are toxic and hazardous to many living cells, and therefore the Ag/AgCl electrode cannot be applied in vivo, or when in vitro cellular vitality is important for the sensing.
In addition to these questionable assumptions in realistic experiments, Ag ions are toxic and cannot be used for in vivo applications. Therefore, although Ag/AgCl reference electrode has still been regarded as the gold standard, many efforts had been applied to decrease the dependence on the reference electrode (Wong & White, 1989) or reduce its invasive impacts (Huang & Huang, 2002).
For electrodes of in vivo measurements, a polarizable (nonfaradic) electrode such as Pt and carbon fiber is often used to give a reference electrostatic voltage to the body fluid. For FET-based sensing purposes, this reference voltage is often the same voltage at the transistor source electrode. Remember that when there is no direct amperometric electrode in touch with the buffer fluid, the fluidic potential will mainly follow the voltage of the largest capacitance, just like the floating-gate equation in Eqs. 12.1 and 12.2. That is, with all polarizable contacts, the buffer behaves like a (leaky) floating gate. Therefore, either we still need a sizable polarizable electrode in the buffer, or we can design a large capacitive coupling from the sensing chip.
Many electrochemical experiments rely on the use of the Ag/AgCl reference electrode for repeatability. Note that the only chemical balance achieved here is the chlorine ion where there is a continuous supply and drain of Cl− during the continuous operation and where the constant bulk an-ion concentration can be maintained. All other cat-ions and an-ions, including even Cl− when Ag/AgCl is depleted or not used, have a constant total dose within the fluidic volume regardless of the variation in surface potentials. For example, if Na+ is pushed away from the surface by a positive surface potential, it would has to pile up in some other places.
Both polarizable and reference electrodes are effective in removing the static electron charge in the electrolyte. Static charge, often due to tribological interaction with the container walls, is well studied in fuels and gasoline to avoid the instantaneous combustion from electrostatic discharge or spark generation. Although such concerns are not needed for most electrolytic buffers, static charge can affect the FET readout by shifting the fluidic potential with respect to the FET source. Therefore, polarizable electrodes can still help stabilize the readout more than pure large capacitive coupling.
3.3 Electrochemical Models for the Fluidic Interface
At the metallic sensing gate of a CνMOS transistor, image charge is induced upon the solid–liquid interaction, and an electrical double layer (EDL) is formed in the liquid side, as illustrated in Fig. 12.7. In the Gouy–Chapman–Stern (GCS) interface model, we have a fixed layer of adsorbed ions (including protons) at the solid–liquid interface which is determined by the chemical functional group on the surface. For example, to consider proton adsorption, the charged hydroxyl group (OH−) is most critical. The charge centroid of this interface layer defines the inner Helmholtz plane (IHP). We can then have a densely packed ion layer when the surface potential is sufficiently high (Kilic et al., 2007), whose outer boundary defines an outer Helmholtz plane (OHP). Due to the steric hindrance (caused by the size of the hydrated ions), the densely packed ion layer can be more than a monolayer but not covalently bonded. Outside OHP, we then have a diffusive layer of free-moving ions (the layer that can be driven in electro-osmotic drive).
According to GCS model (Bard & Faulkner, 2001), the EDL capacitance per unit area can be expressed as
where C H is the Helmholtz-plane capacitance per unit area and can be written as
and the diffuse capacitance per unit area C DIF can be estimated by
where ε buf is the relative permittivity of the solution; ε 0, the permittivity in vacuum; z, the magnitude of charge on the ions; n, the number concentration of the potential-determining ions (Hunter, 2001) in the liquid bulk; k B, the Boltzmann constant; T, the temperature; x OHP, the thickness of the condensed ion layer (Kilic et al., 2007; Storey & Bazant, 2012); and ϕ OHP is the potential at the OHP. The OHP can be envisioned as the closest plane at certain distance x OHP, where the solvated ions can best approach due to the finite size of those nonspecifically adsorbed ions. When a densely packed ion layer is formed under high surface potential which can be physically observed by atomic-force microscopy (AFM) (Siretanu et al., 2014) or reflection ellipsometry (Wang, Zhao, Duits, Mugele, & Siretanu, 2015), the surface is said to reach its steric limit, and further complication of descreening (i.e., ε buf becomes smaller) and charge reversal (i.e., counter-ions may be part of the steric layer) (Landheer, Aers, McKinnon, Deen, & Ranuarez, 2005; Liu et al., 2010) can modify the expression in Eq. 10.8.
The values of ϕ OHP usually range from tens to hundreds of millivolts (Hunter, 2001), and they indicate both the charge condition at the solid–liquid interface and the closest distance the solvated ions can reach (Bard & Faulkner, 2001). The structure of the electrical double layer can affect the interfacial interaction and vice versa, considering the ions in the steric layer is not covalently bonded. Particularly in poly-electrolyte with multiple cat-ions such as Na+, K+, Mg2+, and Ca2+, which are parts of the standard composition in biological buffer solution such as phosphate buffered saline (PBS). Change of ionic composition in the steric layer can be VERY low (on the order of seconds to minutes) due to the dense packing and had been long suspected as one of the reasons for the slow drift in surface potential.
The free-moving ion species can only approach the boundary at OHP, and the ϕ OHP is one of the important factors determining the contribution of the diffuse-layer capacitance C DIF, as in Eq. 12.10, which is often smaller than C H in series and hence set the observable capacitance. Although the relative permittivity of the buffer ε buf is known to be potential dependent as well due to local polarization level (Liu et al., 2010), it is assumed to be constant in our extraction of ϕ OHP for the ease of calculation in the conceptual demonstration.
Note that the sensing gate potential V SG in Fig. 12.2 will be the surface potentialΨ 0 denoted in Fig. 12.7b, which will be a function of the charge composition in the condensed layer and the interface adsorbed charge density σ 0. At high surface potential, as σ 0 is not directly measurable and cannot be pinned down easily (depending on the surface functional group and prior history of acid/salt exposure), we do not have an accurate evaluation of the charge composition in the condensed layer, which renders Ψ 0 hard to predict even we know the bulk concentration. The Gauss law at the interface can be used to evaluate the electric field in the oxide, but we still do not have sufficient information to distinguish Ψ 0 unambiguously. Therefore, differential sensing or V CG compensation to fix V FG is a better way to evaluate the fluid ionic conditions than just using I D to evaluate a variable V FG under different ionic distribution.
3.4 Deposition of a Functional Layer in Postprocessing
Selectivity and sensitivity broadly comprise two most important figures of merit for a sensor, with selectivity often accomplished through careful selection of functional materials in the case of chemical and biological sensing. For chemical sensing, one option is the deposition of a compound such as ionomers or self-assembled monolayers (SAMs) that are selective for one or a group of chemicals (Albert et al., 2000; Freund & Lewis, 1995). Another option is the use of a filter to allow only one type of analyte to reach a broadly sensitive sensor interface. For example, various kinds of insulators have a predictable voltage response to the presentation of ions (van Hal, Eijkel, & Bergveld, 1996). Materials such as SiO2, Si3N4, and Ta2O5 can be grown or deposited on the interface and have predictable Nernst responses of −45, −54, and −58 mV/dec, respectively, with −60 mV/dec being ideal. Ta2O5 functional coating has been used in acidity ISFET sensors for many years.
For biological detection, surface selectivity is usually accomplished by depositing a layer intended to grab a molecule that has high selectivity for the analyte of interest, such as the use of gold/thiol or (3-aminopropyl) triethoxysilane (APTS) chemistry to bind a specific antibody with a given antigen or protein as the target analyte (Jacquot, Muñoz, Branch, & Kan, 2008; Ulman, 1996).
There are many techniques for applying thin functional coating layers to a sensor interface. Techniques for deposition during fabrication include low pressure vapor deposition (LPCVD), plasma enhanced chemical vapor deposition (PECVD), molecular vapor deposition (MCVD), sputtering, and evaporation, which is often not at the CMOS designer’s disposal. Postprocessing steps on the sensing gate surface of the standard CMOS chip include molecular printing, dip coating, dip pen, spin on coating, sol–gel process, SAMs, and aerosol spray. Adhesion and integrity of these postprocessing films are critical for the long-term reliability of the CMOS sensors.
Last but not least, the selectivity through functional groups with very high affinity does cause much confusion in the literature, especially in the case of low bulk concentration and fast sampling when the surface cannot reach equilibrium within the sampling time. Consider the following surface adsorption reaction equation:
where A (l) is the target analyte in the fluid, * is the surface adsorption site defined by the functional group, \( {A}_{\left(\mathrm{s}\right)}^{\ast } \) is the adsorbed target on the surface, and k f and k r are the forward and reverse reaction constants. High affinity of A to the functional group implies k f is very large. The transient response for \( {A}_{\left(\mathrm{s}\right)}^{\ast } \) can be expressed as
When A (l) has sufficient supply in the bulk fluid, the bulk concentration [A (l)] can be regarded as constant, or we will enter into the source depletion mode and Eq. 12.11 cannot readily achieve chemical balance.
The sensor output is a function of the adsorbed ion density and will only be stable when \( {A}_{\left(\mathrm{s}\right)}^{\ast } \) has reached its steady state, which implies:
Note that [A (l)] can only be reliably determined from \( {A}_{\left(\mathrm{s}\right)}^{\ast } \) when the chemical balance of Eq. 12.11 has been achieved. Otherwise \( {A}_{\left(\mathrm{s}\right)}^{\ast}\left( t={t}_1\right) \), and hence the translated [A (l)], will depend on \( {A}_{\left(\mathrm{s}\right)}^{\ast}\left( t={t}_0\right) \) and the sampling period (t 1 − t 0). For very large k f, or equivalently highly selective adsorption with very small k r, chemical balance of Eq. 12.11 can take a long time to achieve especially when [A (l)] is low.
3.5 Transient Models
For transient analyses, consider the time-varying voltage signal at the control gate (ṽ CG) can be expressed as
where V CG0 is the offset DC control-gate voltage and V CG and ω are the magnitude and the angular frequency of the applied AC voltage signal. The sinusoidal AC input has two advantages over DC-only input: (1) AC responses can be treated with simpler small-signal linear models; (2) the output can be filtered at the inquisitive frequency to reduce the noises, especially away from the low-frequency Flicker noise. However, when the test signal period is shorter than the time required to achieve the chemical equilibrium, the surface adsorption of the analyte will stop responding to the AC signal, similar to the cutoff frequency in transistor operations.
Assume the constant drain voltage V D of the MOS transistor is biased at a small value and V S = 0, the above-threshold transient drain current i D is in the linear region and can be expressed by
The induced charge at the solid–liquid interface and the according total net charge Q affects the drain current, which is also influenced by the variation of the total capacitance upon the formation of electrical double layer. In addition, the change of the interface charge may create fluctuation in the potential at the Outer Helmholtz Plane ϕ OHP in the liquid bulk, and cause C SG to change accordingly with the frequency of the AC signal. This effect further influences the total capacitance C T, the C CG/C T ratio on both the swinging amplitude and the offset voltage, and hence the drain current.
The transient measurement is useful in the step response and monotone excitation and sensing. For example, i D can go through a bandpass filter or a lock-in amplifier to investigate only the part with the corresponding frequency, which can be viewed as a generalized impedance spectroscopy with flexibility in the operating point and excitation magnitude. When larger molecules are involved in the EDL, the frequency domain information (magnitude and phase) is valuable to improve specificity (Jayant et al., 2013b).
4 Sample Sensor Measurements
The capacitive load and static charge introduced onto the floating gate cause specific changes in both subthreshold slopes and threshold voltages. The total amount of nonvolatile charge Q on the floating gate further influences the interaction with the fluid at the interface. The drain current from the control-gate voltage sweep then serves as a signature for the fluid and its solute contents. High sensitivity is derived from the high transconductance gain of the MOS transistor and a large dynamic range is achieved concurrently from the large control-gate voltage V CG sweep.
After measuring the full-swing I–V curves for each type of sample fluid in the encapsulated channel, we can also opt to tunnel electrons onto and off the floating gate of the CνMOS device by a fixed bias at the control gate, 30 V for tunneling in and −30 V for tunneling out. The I–V characteristics of various sample fluids are separately recorded after the electron-tunneling operations. The drain voltage is biased at 0.5 V for all the measurements to prevent hot carrier injection. The I–V traces in bare devices and when the microfluidic chamber is filled with deionized water, acetone, and saline are shown in Fig. 12.8. For deionized (DI) water and acetone, the subthreshold slopes change significantly due to the dominant diffuse capacitors in the fluid. The different I–V behavior of water and acetone on polysilicon primarily stems from the smaller ionic strength of acetone. The pH value for DI water is around 5.5 and around 7 for acetone, which is used to estimate the number concentration n in Eq. 12.10. We assume the sampling time is sufficient to build the proton or cat-ion balance at the surface.
The charge interaction further suggests the feasibility of an actuation scheme, in which the amount or distribution of the interfacial charge can be modified in correspondence with the charge tunneled from the control gate into the floating gate. This type of actuation can be utilized in modifying the surface hydrophobicity (e.g., electrowetting (Lee & Kim, 2000; Matsumoto & Colgate, 1990), or in electroosmotic (Kohashi & Kurosawa, 1991; Manz et al., 1994) applications).
The generic SiO2 sensing surface has distinctive response to salinity and acidity from both the contribution of Ψ 0 (which affects the threshold voltage V th) and C DIF (which affects the subthreshold slope S CG). Sample IV sweeps are shown in Figs. 12.9 and 12.10. Note that due to the complicated ion composition in EDL, it is better to use to entire sweep to determine the salinity and acidity than a direct interpretation of I D.
We also monitor the I–V relationship at the transient mode on CνMOS transistors with sample fluids in the microfluidic channels. An AC signal with 5-V amplitude is applied at the control gate, with the frequency ranging from 1 Hz to 100 kHz, as shown in Fig. 12.11. The monitored drain current is amplified and converted to the voltage signal through a transimpedance amplifier with the sensitivity set at 50 μA/V, and the output voltage is then examined on an oscilloscope. At frequencies of 1 Hz–10 kHz the quasi-static behavior has been valid for the control devices as well as the devices with water and saline. Above 10 kHz, significant deviation has been observed, which corresponds to the regime when the ions stop responding to the AC excitation.
5 Summary and Further Reading
Although the fluidic interface has still many complicated electrochemical effects to be resolved before the FET-based sensors can be predictively modeled, there are several sensing modality such as V CG compensation and general impedospectrocopy that can be used today for reliable sensing (Gordon et al., 2015). Research on interface condition and how it can affect sensor reading for acidity, salinity, molecules, action potentials, and vesicle release (Jayant et al., 2013a, 2013b; Jayant, Auluck, Rodriguez, Cao, & Kan, 2014) will still need more efforts for broader sensing applications and controllable actuation. The electronic interface to molecules and cells holds many promises to investigate the secrets of life itself, as at the present time only the electronic devices are up to the task requirements in view of the feasible geometrical scale, the computing/communication system support, and the cost structure. Using the deep integration with CMOS, as we finally have such precise, affordable, and powerful tools after the invention of microscopes, we are at the door of a new research paradigm.
References
Adamson, W. (1982). Physical chemistry of surfaces (4th ed.). New York: Wiley.
Albert, K. J., Lewis, N. S., Schauer, C. L., Sotzing, G. A., Stitzel, S. E., Vaid, T. P., et al. (2000). Cross-reactive chemical sensor arrays. Chemical Reviews, 100(7), 2595–2626.
Allain, J. P. (2000). Clinical & Laboratory Haematology, 22, 1–10.
Bard, J., & Faulkner, L. R. (2001). Double-layer structure and adsorption, Chap. 13. In Electrochemical methods: Fundamentals and applications (2nd ed., pp. 534–577). New York: Wiley.
Berdondini, L., Chiappalone, M., van der Wal, P. D., Imfeld, K., de Rooij, N. F., Koudelka-Hep, M., et al. (2006). A microelectrode array (MEA) integrated with clustering structures for investigating in vitro neurodynamics in confined interconnected sub-populations of neurons. Sensors and Actuators B: Chemical, 114(1), 530–541.
Berdondini, L., van der Wal, P. D., Guenat, O., de Rooij, N. F., Koudelka-Hep, M., Seitz, P., et al. (2005). High-density electrode array for imaging in vitro electrophysiological activity. Biosensors and Bioelectronics, 21, 167–174.
Bergveld, P. (1970). Development of an ion-sensitive solid-state device for neurophysiological measurements. IEEE Transactions on Biomedical Engineering, 17(1), 70–71.
Bousse, L., Shott, J., & Meindl, J. D. (1988). A process for the combined fabrication of ion sensors and CMOS circuits. IEEE Electron Device Letters, 9(1), 44–46.
Burns, J. R., & Powlus, R. A. (1996, July 12). Threshold circuit utilizing field effect transistors. U.S. Patent #3,260,863 (filed March 19, 1964).
Colapicchioni, C., Barbaro, A., Porcelli, F., & Giannini, I. (1991). Immunoenzymatic assay using CHEMFET devices. Sensors and Actuators B: Chemical, 4, 245–250.
Freund, M. S., & Lewis, N. S. (1995). A chemically diverse conducting polymer-based “electronic nose”. Proceedings of the National Academy of Sciences, 92(7), 2652–2656.
Gardner, J. W., Varadan, V. K., & Awadelkarim, O. O. (2001). Microsensors, MEMS, and smart devices. New York: Wiley.
Gordon P. H., Jayant K., Cao Y., Auluck K., Phelps, J. B., & Kan, E. C. (2015). Critical assessment on modeling and design of non-Faradic CMOS electrochemical sensing. IEEE Sensor Journal, art. 2445292.
Huang, I. Y., & Huang, R. S. (2002). Fabrication and characterization of a new planar solid-state reference electrode for ISFET sensors. Thin Solid Films, 406(1–2), 255–261.
Hunter, R. J. (2001). Foundations of colloid science (2nd ed.). New York: Oxford University Press.
Jacquot, B. C., Muñoz, N. L., Branch, D. W., & Kan, E. C. (2008). Non-faradic electrochemical detection of protein interactions by integrated neuromorphic CMOS sensors. Biosensors and Bioelectronics, 23(10), 1503.
Janata, J. (2003). Electrochemical microsensors. Proceedings of the IEEE, 91(6), 864–869.
Jayant, K., Auluck, K., Funke, M., Anwar, S., Phelps, J. B., Gordon, P. H., et al. (2013a). Programmable ion sensitive transistor interfaces I: Electrochemical gating. Physical Review E, 88(1), 012801.
Jayant, K., Auluck, K., Funke, M., Anwar, S., Phelps, J. B., Gordon, P. H., et al. (2013b). Programmable ion sensitive transistor interfaces II: Biomolecular sensing and manipulation. Physical Review E, 88(1), 012802.
Jayant, K., Auluck, K., Rodriguez, S., Cao, Y., & Kan, E. C. (2014). Programmable ion-sensitive transistor interfaces III. Design considerations, signal generation and sensitivity enhancement. Physical Review E, 89(5), 052817.
Jayant, K., Rajwade, S., Pollack, L., & Kan, E. C. (2010) Controlled adsorption and desorption of DNA on CMOS—Towards a bi-directional bio-electronic interface. Biosensors, Galsgow, UK.
Kilic, M. S., Bazant, M. Z., & Ajdari, A. (2007). Steric effects in the dynamics of electrolytes at large applied voltages. I. Double-layer charging. Physical Review E, 75, 021502.
Kohashi, T., & Kurosawa, T. (1991). Electroosmotic display device. IEEE Transactions on Electron Devices, ED-38(9), 2064–2069.
Kovacs, G. T. A., Storment, C. W., Halks-Miller, M., Belczynski, C. R., Jr., Santina, C. C. D., Lewis, E. R., et al. (1994). Silicon-substrate microelectrode arrays for parallel recording of neural activity in peripheral and cranial nerves. IEEE Transactions on Biomedical Engineering, 41(6), 567–577.
Kress-Rogers, E. (1997). Handbook of biosensors and electronic nose: Medicine, food, and the environment. Boca Raton: CRC Press.
Landheer, D., Aers, G., McKinnon, W. R., Deen, M. J., & Ranuarez, J. C. (2005). Model for the field effect from layers of biological macromolecules on the gates of metal-oxide-semiconductor transistors. Journal of Applied Physics, 98, 044701.
Lang, H. P., Baller, M. K., Battiston, F. M., Fritz, J., Berger, R., Ramseyer, J.-P., et al. (1999). The nanomechanical NOSE. In Proceedings of the IEEE Micro Electro Mechanical Systems (MEMS ‘99), pp. 9–13.
Lee, J., & Kim, C.-J. (2000). Surface-tension-driven microactuation based on continuous electrowetting. Journal of Microelectromechanical Systems, 9(2), 171–180.
Liu, Y., Huber, D. E., Tabard-Cossa, V., & Dutton, R. W. (2010). Descreening of field effect in electrically gated nanopores. Applied Physics Letters, 97, 143109.
Lonergan, M. C., Severin, E. J., Doleman, B. J., Beaber, S. A., Grubbs, R. H., & Lewis, N. S. (1996). Array-based sensing using chemically sensitive, carbon black-polymer resistors. Chemistry of Materials, 8, 2298–2312.
Madou, M. J., & Morrison, S. R. (1989). Chemical sensing with solid state devices. San Diego: Academic.
Manz, A., Effenhauser, C. S., Burggraf, N., Harrison, D. J., Seiler, K., & Fluri, K. (1994). Electroosmotic pumping and electrophoretic separations for miniaturized chemical analysis systems. Journal of Micromechanics and Microengineering, 4, 257–265.
Matsumoto, H., & Colgate, J. E. (1990) Preliminary investigation of micropumping based on electrical control of interfacial tension. In Proceedings of the IEEE Micro Electro Mechanical Systems (MEMS), pp. 105–110.
Minch, B. A. (1997). The subthreshold floating-gate MOS transistor, Chap. 5. Ph.D. Dissertation, California Institute of Technology, Pasadena, CA, pp. 127–158.
Shen, Y. N., Liu, Z., Jacquot, B. C., Minch, B. A., & Kan, E. C. (2004). Integration of chemical sensing and electrowetting actuation on chemoreceptive neuron MOS transistors (CνMOS). Sensors and Actuators B, 102(1), 35–43.
Shen, N. Y., Liu, Z., Lee, C., Minch, B. A., & Kan, E. C. (2003). Charge-based chemical sensors: A neuromorphic approach with chemoreceptive neuron MOS (CνMOS) transistors. IEEE Transactions on Electron Devices, ED-50(10), 2171–2178.
Shibata, T., & Ohmi, T. (1992). A functional MOS transistor featuring gate-level weighted sum and threshold operations. IEEE Transactions on Electron Devices, ED-39(6), 1444–1455.
Siretanu, I., Ebeling, D., Andersson, M. P., Stipp, S. L. S., Philipse, A., & Stuart, M. C. (2014). Direct observation of ionic structure at solid-liquid interfaces: A deep look into the Stern Layer. Scientific Reports, 4, 4956.
Smith, R., Huber, R. J., & Janata, J. (1984). Electrostatically protected ion sensitive field effect transistors. Sensors and Actuators, 5(2), 127–136.
Steiner, F. -P., Hierlemann, A., Cornila, C., Noetzel, G., Bachtold, M., Korvink, J. G., et al. (1995). Polymer coated capacitive microintegrated gas sensor. In Proceedings of the 8th International Conference on Solid-State Sensors and Actuators (Transducers ‘95), pp. 814–817.
Stern, E., Steenblock, E. R., Reed, M. A., & Fahmy, T. M. (2008). Label-free electronic detection of the antigen-specific T-cell immune response. Nano Letters, 8(10), 3310–3314.
Storey, B. D., & Bazant, M. Z. (2012). Effects of electrostatic correlations on electrokinetic phenomena. Physical Review E, 86, 056303.
Ulman, A. (1996). Formation and structure of self-assembled monolayers. Chemical Reviews, 96(4), 1533–1554.
van der Spiegel, J., Lauks, I., Chan, P., & Babic, D. (1983). The extended gate chemically sensitive field effect transistor as multi-species microprobe. Sensors and Actuators, 4, 291–298.
van Hal, R. E. G., Eijkel, J. C. T., & Bergveld, P. (1996). General model to describe the electrostatic potential at electrolyte oxide interfaces. Advances in Colloid and Interface Science, 69, 31–62.
Wang, L., Zhao, C., Duits, M. H. G., Mugele, F., & Siretanu, I. (2015). Detection of ion adsorption at solid–liquid interfaces using internal reflection ellipsometry. Sensors and Actuators B: Chemical, 210, 649–655.
Wenzel, S. W., & White, R. M. (1988). A multisensor employing an ultrasonic lamb-wave oscillator. IEEE Transactions on Electron Devices, ED-35(6), 735–743.
Wong, H. S., & White, M. H. (1989). A CMOS-integrated ISFET-operational amplifier chemical sensor employing differential sensing. IEEE Transactions on Electron Devices, 36(3), 479–487.
Yager, P., Domingo, G. J., & Gerdes, J. (2008). Annual Review of Biomedical Engineering, 10, 107–144.
Author information
Authors and Affiliations
Rights and permissions
Copyright information
© 2017 Springer International Publishing AG
About this chapter
Cite this chapter
Ma, Y., Kan, E. (2017). CMOS Biosensors. In: Non-logic Devices in Logic Processes. Springer, Cham. https://doi.org/10.1007/978-3-319-48339-9_12
Download citation
DOI: https://doi.org/10.1007/978-3-319-48339-9_12
Published:
Publisher Name: Springer, Cham
Print ISBN: 978-3-319-48337-5
Online ISBN: 978-3-319-48339-9
eBook Packages: EngineeringEngineering (R0)