A CMOS Biocompatible Charge Detector for Biosensing ... - IEEE Xplore

2512

IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 59, NO. 9, SEPTEMBER 2012

A CMOS Biocompatible Charge Detector for Biosensing Applications Stefano Lai, Alessandra Caboni, Daniela Loi, and Massimo Barbaro, Member, IEEE

Abstract—A solid-state CMOS device capable of detecting the changes of electric charge caused by a chemical or biological reaction is presented. The device is fully compatible with a standard CMOS process. Biocompatibility is obtained by the passivation of the active area with a layer of alumina obtained with a simple, low-cost, and reliable process. A test chip hosting 80 sensors has been realized and characterized showing the detection capabilities of the novel sensor. The reusability of the device by stripping of the alumina layer was proved. Index Terms—Biosensors, CMOS process, double-gate fieldeffect transistors (FETs).

I. I NTRODUCTION

M

ANY IMPORTANT biological and chemical reactions entail an electric charge variation. If such modification is localized and reproducible, it can be detected by a fieldeffect sensor. Such kind of devices, when realized in a CMOS process, benefits from all the advantages provided by solid-state electronics: a mature and scalable technology, large integration, and, consequently, low fabrication cost for a single unit. For these reasons, integrated field-effect transistor (FET)-based sensors seem to be a promising approach for the realization of simple portable inexpensive detection platforms which will hopefully substitute the expensive, bulky, and complex instrumentation currently used for several analyses in pharmaceutical and medical laboratories. In the last decade, a lot of solid-state field-effect electric charge sensors have been proposed for several application fields. Charge-sensitive devices have been applied to the detection of chemical reactions [1]; chemical FET-based sensors have been developed for gas analysis [2]. Charge detection is widely exploited for biomolecular analysis: Variation in the environmental charge amount can be linked to different biological phenomena, including enzymatic reactions [3], antigene–antibody interaction [4], and cellular activity [5]. Genetic analysis represents a promising application field for the so-called BioFET; in the literature, a lot of examples are reported, including single devices [6] and integrated solutions [7], [8]. Manuscript received May 26, 2011; revised November 7, 2011, January 18, 2012, March 27, 2012, and May 24, 2012; accepted May 24, 2012. Date of publication June 26, 2012; date of current version August 17, 2012. The review of this paper was arranged by Editor D. Verret. The authors are with the Department of Electrical and Electronic Engineering, University of Cagliari, 09123 Cagliari, Italy (e-mail: stefano.lai@ diee.unica.it; [email protected]; [email protected]; [email protected]). Color versions of one or more of the figures in this paper are available online at http://ieeexplore.ieee.org. Digital Object Identifier 10.1109/TED.2012.2202233

Such devices usually are not fully compatible with a standard CMOS process and require the use of sophisticated structures, like microelectrochemical cells and electrodes, or non-CMOS materials (gold, silver, chrome, etc.); this determines a cost increase and the reduction of flexibility of the technology. In particular, direct interface with standard electronic devices becomes difficult and complex. Nonstandard materials are often necessary, depending on the application field, particularly if biocompatibility is required. Moreover, the detection is often indirect, and what is measured is not the reaction itself but a signal coming from a label or modification in the original molecule. To overcome all these limitations, a new approach has been recently proposed for field-effect chemo- and biosensors [9], [10]: The charge associated to a molecule immobilized on a sensing area is directly detected and evaluated by a floatinggate FET. This kind of structure can be realized in a standard CMOS process: The sensing area is an exposed section of the floating gate, covered by the native aluminum oxide (Al2 O3 ), as proposed in [11], or by a layer of silica (SiO2 ) introduced in the realization process [12]. The immobilization of biomolecules on these biocompatible materials has been investigated in [13]: Silica demonstrates a higher grade of immobilization, particularly because of the low quality and regularity of the native alumina. On the other hand, direct immobilization of molecules on silica shows some disadvantages: First of all, the silica works as a thick spacer, reducing the efficiency of the field effect; second, reuse of the sensor is difficult since eliminating the molecules requires chemical etching of silica, which represents a disadvantage, particularly in the prototyping phase. In this paper, a new approach to the solution described in [11] is proposed. The charge sensor hosts a metallic sensing area, which undergoes a simple low-cost oxidation treatment in order to obtain an alumina layer with superior quality in terms of uniformity, thickness, and robustness in the aqueous environment typically used in chemical and biological analyses. The alumina layer is thinner than silica layers in standard commercial CMOS processes and can be also removed with chemical etching without significantly affecting the aluminum surface; this feature allows the reusability of the sensor for further uses.

II. S ENSING D EVICE A. Device Structure The sensing unit is based on the charge-modulated FET, whose working principle has been described, modeled, and

0018-9383/$31.00 © 2012 IEEE

LAI et al.: CMOS BIOCOMPATIBLE CHARGE DETECTOR FOR BIOSENSING APPLICATIONS

Fig. 1.

2513

(a) Sensing unit cross section, (b) equivalent circuitry, and (c) structure of the sensing site.

simulated in [9]. Such principle was validated with the fabrication of a test chip [14] in a standard 0.8-μm CMOS process from Austriamicrosystems. Successful experimental results demonstrated the possibility of applying this charge sensing technique to label-free detection of DNA hybridization. The first prototype chip hosted 16 sensory sites, and all of them were NMOS based. This choice limited the range of detectable negative charge since detection was based on the measure of the changes in the threshold voltage: An excessive increase of the threshold largely decreases output currents. As a consequence, the readout process becomes slower, SNR increases, and, eventually, the transistors shut off. The same would be true for positive charges with a sensor based on a PMOS device. To address this problem, the sensing unit was modified in a PMOS–NMOS pair. In this way, when the sensing charge is negative and the NMOS enters the cutoff region, the PMOS conducts even larger currents providing extended detectable ranges. Vice versa, when the electric charge is positive and the PMOS shuts off, the NMOS is still capable of detection. For small amounts of positive or negative charge, both the devices conduct, and a combination of the two outputs can be exploited to get a redundant information. A scheme of the novel sensing unit is shown in Fig. 1(a). The control gate and the floating gate realize a plane linear poly1–poly2 capacitor (CCF ), whose capacitance depends upon the oxide thickness and gate area. In this way, acting on the control gate, it is possible to set the operating point of the device by capacitive coupling. By properly choosing the control-gate voltage, it is possible to nullify the effect of the unknown charges trapped in the floating plane during the manufacturing process. The double poly capacitor is covered with metal layers alternately set at the floating- and control-gate voltages. This technique allows to increase the overall capacitance and to reduce the potential crosstalk between the floating gate and other signal wires. An additional grounded metal layer, not shown in the cross section, was used to shield the sensing transistors from undesired field effects. The control capacitors have three possible dimensions, shown in Fig. 2(b). The sensing area consists in a pad opening in the Si3 N4 passivation layer that allows to access a metal pad on which the target molecules can be anchored. Such pad is electrically connected to the floating gate as shown in Fig. 2(c). Three different dimensions are used for such pads, as shown in Fig. 2(b). With respect to [14], the sensing pad

Fig. 2. (a) Microphotograph of the chip and (b) a detail of the sensors.

presents two additional electrodes, placed at both sides of the pad opening, and that can be independently forced at supply voltage, ground, or left floating. Such electrodes are used to provide the possibility of generating a transversal electric field which could speed up or induce specific chemical reactions such as DNA hybridization; their presence must be taken into account when modeling the response of the sensor. B. Device Model In order to extrapolate the amount of immobilized electric charge from measures of the electrical parameters, an analytical model of the device was developed. The model focuses particularly on how the parasitic capacitors and their interconnections influence the sensor output. The main parasitic capacitors are shown in Fig. 1(b). The capacitor between the sensing pad and the n-well, CPAD , represents the most important parasitic contribution in the model; its value can be found with the relationship CPAD = Ca · A +

5

Cf 1,i · (2l1 + l2 )

(1)

i=1

where Ca is the capacitance per area, A is the area of the pad, Cf 1,i is the fringing capacitance to the underlying conductor in case of adjacent lines for the ith metal layer ([F/m]), and l1 and l2 are the dimensions of the electrodes as reported in Fig. 1(c). In a similar way, it is possible to obtain the CCF capacitor between the control and floating gates. Another contribution is given by the horizontal capacitors between the sensing pad and

2514


electrodes [see Fig. 1(c)], Clelect and Crelect ; this capacitance is equal for the right and left electrodes, and its value can be found with the relationship Crelect = Clelect =

5

Cc,i · (2l1 + l2 )

(2)

i=1

where Cc,i is the coupling capacitance for adjacent lines for the ith metal layer ([F/m]). Finally, the parasitic capacitors of NMOS and PMOS have to be considered. For each transistor, the main contribution is given by the overlap capacitors between the floating gate and source (CFS ) and the floating gate and drain (CFD ) and by the nonlinear capacitance between the floating gate and bulk (CFB ). Such capacitors can be evaluated from spice analysis. In operating conditions, the transistors work in saturation region, so the floating-gate–drain capacitance is negligible if compared to other contributions and is excluded from the model. Applying the charge conservation principle to the floating gate and solving the resulting equation with respect to VFG , this relationship can be found VFG =

CCF + CFS,N + CFS,P Crelect VCG + Vrelect CTOT CTOT +

Clelect CPAD + CFB,P Vlelect + VDD CTOT CTOT

+−

CFS,N CFS,P QTOT ΔVN + ΔVP + CTOT CTOT CTOT

(3)

where Vlelect and Vrelect are the voltages of the left and right electrodes, respectively, ΔVN and ΔVP are the overdrive voltages for PMOS and NMOS, CTOT is the sum of all the capacitive terms, and QTOT is the total charge on the floating gate. Before immobilization of the charge to be sensed, it coincides with the native charge Q0 . The relationship is linear in VCG , so (3) can be rewritten as VFG = m · VCG + q.

(4)

The contribution of capacitive terms can be taken into account estimating the slope and intercept of (4). In particular, the intercept contains information on the immobilized charge Qi and other terms that are unknown (the native charge) or difficult to evaluate a priori (ΔVN and ΔVP ). In order to simplify the estimation of charge without considering these terms, a correlated double-sampling technique can be adopted: If ΔVFG is the variation of the floating-gate voltage after the immobilization of charged molecules on the active area, QTOT changes from Q0 to Q0 + Qi , and this last term can be extracted as Qi = CTOT · ΔVFG .

(5)

For a given Qi , the smaller is CTOT , the larger is ΔVFG and thus, apparently, the sensitivity. Anyhow, CTOT depends on CPAD and CCF , so there are a number of drawbacks in its decrease. Reducing CPAD (i.e., reducing the area of the sensing pad), the amount of available charge is consequently reduced. Moreover, CCF cannot be reduced arbitrarily but must be larger than the sum of all the other capacitance in the layout in order to properly bias the devices.

C. Test Chip The previously described sensing structure was embedded in a test chip, implemented in a standard 0.35-μm CMOS process with two poly and five metal layers from AMI Semiconductors (now ON Semiconductors). The chip measures 5 mm2 , including pads. A microphotograph of the chip is shown in Fig. 2(a). The 80 sensors are equally divided into two channels in order to allow differential measurements. The sensing devices belong to seven different typologies, given by different combinations of the sizes of the control capacitor and of the sensing pad reported in Fig. 2(b), where a detail of the channel is shown: The control capacitors and the sensing pad are visible, while other electrical structures are masked by the metal fills. III. R EALIZATION OF THE B IOCOMPATIBLE I NTERFACE In order to correctly immobilize biological material (biomolecules and cells) on the sensing area, a spacing layer has to be realized. For its chemical and physical characteristics, aluminum oxide can be used as the spacer and cross-linking element between the sensing area and the molecules: Its oxide is very reactive, and several kinds of methods can be used for anchoring molecules to its surface [15]. Moreover, aluminum is toxic and simply corruptible in solutions, so the aluminum oxide is important to protect the pad area and to guarantee the biocompatibility of the sensor [16]. An aluminum surface naturally oxidizes, developing a poorly uniform thin layer that easily incorporates dirt and residuals that are likely to compromise the sensing operations. For several applications, a controlled grown aluminum oxide has to be realized. Different techniques can be used for the oxidation process: Aluminum anodization with electrochemical cells [17] and different kinds of plasma [18] are the most common methods to obtain aluminum oxide with specific characteristics of robustness, roughness, uniformity, and thickness. Ozone is another oxidizing agent that could be used in aluminum oxidation processes; aluminum exposed to ultrapure ozone in ultrahigh vacuum chambers develops an oxide layer with improved characteristics of regularity, thickness, and robustness with respect to other kinds of controlled grown oxides [19]. Ozone can be also produced by means of a UV lamp working at a proper wavelength (around 185 nm), generally used for surface sterilization and activation. This kind of instrumentation can be used for a simple low-cost oxidation of metallic surfaces at room temperature and air pressure, with an obvious limitation in the quality of the realized oxide in respect to that obtained by means of the mentioned sophisticated setups. In this paper, two different oxidizing techniques, i.e., air plasma and UV-produced ozone, have been investigated; both techniques have been compared with the use of native aluminum oxide. In particular, the techniques have been compared with respect to the processes necessary to bind single-stranded DNA oligonucleotides on the sensing pads, thus realizing a probe layer necessary for DNA hybridization detection. A. Electrical Characterization In order to electrically characterize the different kinds of investigated aluminum oxides, test capacitors were realized and


2515

TABLE I E LECTRICAL P ROPERTIES FOR THE D IFFERENT I NVESTIGATED A LUMINUM OXIDES

voltage of the system (VDD = 3.3 V); this makes the native aluminum oxide a weak candidate for adoption on an active chip. B. Biocompatible Layer

Fig. 3. Ozone-obtained aluminum oxide thickness extracted from capacitive measurements. In the insert, the structure of a single capacitor is reported.

measured. For this characterization, the nature of the realized oxide is the only parameter of interest, so the results are readily applicable also to the alumina obtained with the same technique on the manufactured chips. The structures are the ones shown in the insert in Fig. 3: The bottom plate is realized in aluminum thermally evaporated on a plastic substrate (polyethylene terephthalate), while the top plate is realized by photolithography of a thermally evaporated gold surface; the average area of each capacitor is 3 × 0.3 mm2 . Three different substrates, hosting 12 capacitors each, were realized, one for each kind of oxide. The native oxide was obtained by means of 90 ◦ C baking of the substrate on a hot plate for 12 h at free air. The air-plasma oxidation was performed with a March Plasmond System, at room temperature, at a pressure of 0.5 torr and with an RF power of 100 W and an oxidation time of 10 min [18]. The ozone oxidation was performed exposing the aluminum surfaces to a UV lamp (UVP Penray) at room temperature and at free air. In order to choose a proper ozone oxidation time, the process was calibrated extrapolating the oxide thickness by capacitive measures. In Fig. 3, the extracted thickness is shown, altogether with the standard error, which depends on photolithography and instrumentation uncertainty. The exposure times are chosen according to the reproducibility of the capacitor realization process: Under 540 s, it is difficult to obtain test capacitors because of poor aluminum resistance to the chemical etching used during photolithography and because of short circuits between the two plates, a symptom of nonuniform oxidation. Fig. 3 shows a saturation trend of the oxidation process and demonstrates the possibility to obtain a uniform aluminum oxide, with good insulation properties (resistance of the capacitors was found to be on the order of megohms) in the nanometric scale. For this reason, in further tests, the oxidation time of 10 min (the same as that of the plasma technique) was adopted. In Table I, the average results of the capacitive analysis are summarized. From the electrical point of view, the three investigated techniques seem to produce quite similar alumina layers: The three different oxides present similar capacitance and leakage currents. As the only remarkable difference, the breakdown voltage of the native oxide is lower than the supply

Aside from the electrical characterization, the different kinds of aluminum oxides were characterized in respect to a case study application in order to verify if the realized layer was suited to resist the chemical and biological procedures necessary to bind molecules of interest on its surface and guarantee the biocompatibility of the whole system. The tests focused on the application of the chip as DNA hybridization sensor, which requires the anchoring of single-stranded DNA oligonucleotides (probes) on the surface of the sensing areas (functionalization). Correct binding of the probes strongly depends upon the characteristics of the oxide layer. The functionalization steps are the same as that used in [11], previously described in [20]. The 31-base single-stranded DNA probes are modified with a thiol group (HS), necessary to link them to the surface, and with a fluorescent dye (Cyanine 3) which emits in the band of yellow green. The thiol group cannot be used for the direct binding of the probes to the alumina surface: A 3-mercaptopropyltrimethoxysilane (3-MPTS) selfassembled monolayer is used as cross-linker layer between the oxide surface and the thiol group. Both the deposition of the 3-MPTS layer (silanization) and the functionalization are realized by the drop casting of the solutions on the test areas. Such steps were applied sequentially on a single prototype chip. After each step, fluorescence was evaluated in order to estimate the efficiency of immobilization. Then, the DNA molecules and intermediate layers were stripped, and the sequence was repeated using a different oxidation technique. In this way, it was possible to compare the three oxidizing techniques and, at the same time, verify the reusability of the device. The analysis started with the native oxide layer which is the one normally present on the surface of the sensing pads of the chip after manufacturing; all the further steps were realized on it without any treatment (except cleaning). As regards the plasma and the ozone oxidation, the procedures described in Section III-A could be applied only after the removal of the native oxide and exposure of the pure aluminum surface. In this way, it was possible to obtain a high-quality oxide layer and also eliminate residuals incorporated in the native oxide that could pollute the environment in which the (bio)molecules are anchored thus reducing the sensitivity. Nonoxidized aluminum can be obtained exposing the oxidized-aluminum surface at a dilution of sulphuric acid in deionized water. The time of exposure and the concentration of sulphuric acid were investigated: A 1:100 solution of sulphuric acid in deionized

2516


Fig. 4. Fluorescence analysis on the surface of the sensing areas of the chip: (a) Not processed chip; (b) silanized chip (over native alumina); (c) functionalized chip with native alumina; (d) functionalized chip with plasma-grown alumina; (e) functionalized chip with ozone-grown alumina; (f) chemical etching (1:100 in DI water) of an ozoned, silanized, and functionalized chip; and (g) green, red, and yellow components of the fluorescent signal coming from the sensing pads of the chip.

water and 10 min of exposure were found to be the best parameters in order to easily reproduce the cleaning technique without compromising the sensing structure. The same procedure was adopted to remove, altogether with the aluminum oxide, the silane and oligonucleotides layers; this process (referred as stripping) permits to completely restore the original surface, so the chip can be reoxidized and reused for further analyses. The results are shown in Fig. 4 which depicts the images of the sensing areas of the chip taken at the fluorescent microscope after the silanization and functionalization processes. The excitation wavelength has its peak at 550 nm (visible green) while the Cy3-modified oligonucleotides have an emission spectrum with wavelengths peaking at 570 nm, i.e., shifted at the border between green and yellow. For this reason, not functionalized surfaces should appear as green (since, in this case, the signal comes from the backscattered light of the excitation source) while functionalized surfaces should appear green yellow: The higher the yellow component, the higher the number of fluorochromes and, thus, the better the immobilization of probe molecules. The image of the aluminum surfaces of a not processed chip (a) is prevalently green as expected and as confirmed in Fig. 4(g). In Fig. 4(b), the image of a silanized native alumina surface is shown; the 3-MPTS layer appears to strongly reduce the autofluorescence of the surface. From Fig. 4(c) to 4(e) show the functionalization of the different kinds of oxide surfaces. If compared to the silanized surface, all the surfaces are apparently functionalized. Anyway, an increasing amount of yellow fluorescence is evident going from native oxide to plasma- and ozone-grown oxides. This evidence is well confirmed by the analysis of the fluorescent signal in Fig. 4(g). Interestingly enough, in the case of ozone-grown alumina, there is a strong contrast between the sensing pads and the surrounding area (covered by the Si3 N4 standard passivation layer) which confirms that the immobilization of molecules is much more efficient over alumina than over silicon nitride. This evidence shows a good selectivity of this kind of chemistry with respect to the surface. Anyhow, this is not a key factor since molecules that may be adsorbed or anchored outside of the pads do not affect the sensitivity of the device. The overall distance between the passivated surface and silicon body, in

fact, is around 11 μm which is enough (altogether with CMOS process features such as field oxide or field implants and design choices such as the use of shielding structures) to guarantee that no field effect can be generated. Finally, Fig. 4(f) shows an active area after stripping of the functionalized layers: The sensing pad is green again, proving the complete removal of the 3-MPTS layer and, thus, of the anchored oligonucleotides. It is important to point out that the proposed stripping technique does not depend upon the kind of chemistry used for anchoring the molecules on alumina, with all the molecular layers being removed altogether with the alumina. Moreover, particularly in the case of DNA analyses, such a procedure could be more efficient than others (e.g., breaking of the thiol bond of the oligonucleotides on the 3-MPTS layer) in terms of preventing contamination between different analyses on the same reusable chip. The stripping procedure was repeated on several chips, used as DNA sensors, and each one could be reused at least four times before an evident deterioration of the aluminum pad was visible. Such results are coherent with the chemical and morphological properties of the different kinds of alumina surfaces. Plasma and ozone treatments are well known to favor the formation of hydroxyl (-OH) groups on the surface of oxides [18] where the 3-MPTS binds; this explains the large difference of fluorescence of DNA molecules bound on 3-MPTS layer realized on native oxide with respect to that realized on plasma- and ozone-grown oxides. The residual difference can be explained by means of atomic force microscopy analyses of their surfaces [see Fig. 4(g)]. The density of grain boundaries was proved to highly influence the uniformity and the stability of SAMs [21]: The plasma-oxidized surface appears less uniform with respect to the ozoned one, so a lower quality of the 3-MPTS layer should be expected and, thus, a lower efficiency in the binding of the oligonucleotides on its surface.

IV. E XPERIMENTAL R ESULTS A. Experimental Setup The floating-gate voltage (VFG ) of each sensing unit was extrapolated with the readout circuit (for the PMOS device)


2517

reason for such differences. In general, the differences between extracted and theoretical values are due to secondary capacitive terms neglected in the model (e.g., layout capacitances), whose effect explains why the measured slopes change accordingly with the connection of the electrodes instead of staying constant as predicted by the model. C. Charge Sensitivity

Fig. 5.

Readout circuit: Level shifter.

shown in Fig. 5(a), a level shifter integrated on chip; for the NMOS sensor, the complementary circuit was used. The circuit is biased with two programmable currents I1 and I2 , and the output voltage Vout is taken at the drain terminal of a diodeconnected PMOS transistor, matched with the sensor. The two currents I1 and I2 are generated by means of two programmable current generators, whereas the control gate of the sensor is driven by a D/A converter. Using the square-law characteristic of a MOS transistor in saturation and noting that the current flowing in the sensor is Isens = I1 − I2 , (6) can be derived Vout = VFG + k( I1 − I2 − I2 ) (6) where k is a constant dependent on process and design parameters. The programmability of bias voltage Vin and bias currents allows to change the linear range of the circuit, which is limited by the minimum voltage drop required to keep all transistors in saturation. Substituting (4) in (6) and choosing I1 = 2I2 Vout = mVin + q.

(7)

Therefore, the circuit allows to plot (4) by applying a ramp to voltage Vin (which corresponds to VCG ). The curves saturate when the transistors enter in the triode or cutoff region. The linear portion of the curve allows to extrapolate parameters m and q and, thus, the immobilized charge Qi . Finally, (5) can be expressed in terms of output voltage as Qi = CTOT · ΔVout .

(8)

B. Model Validation Table II shows the extracted average slopes for the different sensing structures (μm ), the standard deviation (σm ), and the theoretical value extracted from the model (mth). These values are considered for different connections of the electrodes surrounding the sensing pad: both the electrodes connected to ground or to supply voltage VDD , and one electrode connected to ground and the other to VDD . The model proposed is validated with a very small percentage of error () for all the sensing structures except for small control capacitor–medium sensing pad typologies (SC-MP); in this case, the small number of structures on which the average is calculated could be the

In order to evaluate charge sensitivity, the sensor’s response to the oxidation process was measured. During the aluminum oxide’s growth, negative electrical charges are trapped on it [22] and could be evaluated by the sensing units. After the oxidation process, in order to verify the electrical nature of the evaluated charge, a chemical etching was carried out, thus removing the aluminum oxide and resetting the charge. Fig. 6(a) shows the output voltage of a PMOS connected to the level shifter, in the configuration shown in Fig. 5. Successive oxidation steps cause the shift of the output voltage while leaving the slope unchanged [see (3)]. The charge accumulated in the oxide determines the shift in the intercept of the regression curve. After chemical etching, the charge is removed, thus resetting the output voltage to the original value; the chemical etching is not perfectly controlled, so a perfect charge deletion is quite difficult to obtain. In order to evaluate the minimum amount of charge which determines an evaluable variation in the threshold voltage of the transistor, the shifts in the output voltages of the small-padded sensor were considered. With the smallest pad, in fact, for a given oxidation time, the smallest amount of charge is trapped. Fig. 6(b) shows the average output voltage shift (for a given Vin ) and the correspondent extrapolated charge versus the oxidation time. It is possible to notice that even very short oxidation times (down to 1 s) are sufficient to obtain a sensible variation. As expected, the shift is smaller for the small-padded sensors with medium control capacitors that, with CTOT being higher, have the lowest sensitivity; for this kind of structures, the obtained shift in the output voltage is 3.8 mV. Knowing that, for these sensors, CTOT = 2.8 pF, (8) allows to directly extrapolate the corresponding amount of charge variation which is 1.1 · 10−14 C. In the application case of the study of DNA detection, since each oligonucleotide bears an electric charge equal to N times the electron charge q (N being the number of bases), the minimum number of detectable molecules is 1433, around six molecules per square micrometer. The maximum detectable charge can be estimated considering a shift in the threshold voltage which determines the saturation of the output curves of Fig. 5. At most, this shift can be assumed equal to VDD , so considering the medium-control-capacitor big-padded sensors which achieve the best tradeoff between the amount of immobilized charge and sensitivity, a maximum charge of 9.16 · 10−11 C is obtained. In Fig. 7(a), the average charge variation is shown for the different dimensions of the sensing pad. The entity of the variation is clearly dependent on the area of the sensing pad, and a saturation of the oxidation process similar to the one verified in Fig. 3 is detected. The initial charge, set as zero, could be recreated with a good precision for small-padded and middle-padded structures; for the large-padded sensors, the chemical etching probably removes a thicker layer of aluminum

2518


TABLE II T HEORETICAL S LOPE (mth), E XTRACTED AVERAGE S LOPE (μm ), S TANDARD D EVIATION (σm ), AND R ELATIVE D IFFERENCE () FOR T HREE D IFFERENT E LECTRODE C ONFIGURATIONS : (A) B OTH E LECTRODES C ONNECTED TO G ROUND , (B) B OTH E LECTRODES C ONNECTED TO S UPPLY VOLTAGE , AND (C) O NE E LECTRODE C ONNECTED TO G ROUND AND O NE E LECTRODE C ONNECTED TO S UPPLY VOLTAGE

Fig. 6. (a) Output voltages of a PMOS during the oxidation process. (b) Average threshold voltage shift and charge variation detected from the small-padded sensors, with small and medium control capacitors.

oxide with respect to the one grown during the process, so the final charge is larger than the initial one. It is important to point out that PMOS and NMOS on the sensing unit verify the same charge variation. Finally, in Fig. 7(b), the overall oxidation process is evaluated. The charge extracted by every single sensor on a channel is normalized by the sensing pad’s area, so the average charge per area unit can be extracted above the entire channel. In this figure, the outline of the charge trapping process (from “0 s” to “540 s”) is clearly similar to the one shown in Fig. 3 for the aluminum oxide’s growth, proving the relationship between the evaluated charge and the oxide production’s process. V. C ONCLUSION A solid-state device for the detection of electric charge modifications caused by chemical or biological reactions was presented. The device can be implemented in a standard CMOS process. Unlike other applications, activation of the sensing

Fig. 7. (a) Average charge variation during the oxidation process versus sensing area’s dimension. (b) Average charge variation per area during the oxidation process.

area is possible through a simple, low-cost, and reliable postprocessing. As a major improvement with respect to similar CMOS-based applications for biological analyses, the proposed process allows to realize a biocompatible interface that is completely removable by means of chemical etching, so the chip can be reusable. A test chip hosting 80 sensors was realized, and experimental results were provided to show the possibilities of the approach and of the activation process. The reusability of the chip was proved after the stripping of the alumina interface layer. The chip was successfully tested as charge sensor; the range of detectable charges goes from


around 10−14 C to around 10−10 C, thus spanning over four orders of magnitude. This includes the typical range of charges associated with molecular and chemical reactions.

ACKNOWLEDGMENT The authors would like to thank Prof. G.-L. Ferri and Dr. P. Cosseddu for their help with fluorescence analyses and AFM surface analyses. S. Lai gratefully acknowledges Sardinia Regional Government for the financial support of his Ph.D. scholarship.

2519

[18] A. N. Rider, R. N. Lamb, and M. H. Koch, “Low-power RF plasma oxidation of aluminium,” Surf. Interface Anal., vol. 31, no. 4, pp. 302– 312, Apr. 2001. [19] A. Kuznetsova, J. T. Yates, Jr., G. Zhou, J. C. Yang, and X. Chen, “Making a superior oxide corrosion passivation layer on aluminum using ozone,” Langmuir, vol. 17, no. 7, pp. 2146–2152, Apr. 2001. [20] A. Alessandrini, V. De Renzi, L. Berti, I. Barák, and P. Facci, “Chemically homogeneous, silylated surface for effective DNA binding and hybridization,” Surf. Sci., vol. 582, no. 1–3, pp. 202–208, May 2005. [21] S. E. Creager, L. A. Hockett, and G. K. Rowe, “Consequences of microscopic surface roughness for molecular self-assembly,” Langmuir, vol. 8, no. 3, pp. 854–861, Mar. 1992. [22] J. H. Lee, K. Koh, N. I. Lee, M. H. Cho, Y. K. Ki, J. S. Jeon, K. H. Cho, H. S. Shin, M. H. Kim, K. Fujihara, H. K. Kang, and J. T. Moon, “Effect of polysilicon gate on the flatband voltage shift and mobility degradation for ALD- Al2 O3 gate dielectric,” in IEDM Tech. Dig., Dec. 2000, pp. 645–648.

R EFERENCES [1] R. E. G. van Hal, J. C. T. Eijkel, and P. Bergveld, “A novel description of ISFET sensitivity with the buffer capacity and double-layer capacitance as key parameters,” Sens. Actuators B, Chem., vol. 24, no. 1–3, pp. 201–205, Mar. 1995. [2] J. A. Covington, J. W. Gardner, D. Briand, and N. F. de Rooij, “A polymer gate FET sensor array for detecting organic vapours,” Sens. Actuators B, Chem., vol. 77, no. 1/2, pp. 155–162, Jun. 2001. [3] J. Janata and S. Caras, “Field effect transistor sensitive to penicillin,” Anal. Chem., vol. 57, no. 9, pp. 1920–1923, Aug. 1985. [4] M. Gotoh, M. Suzuki, I. Kubo, E. Tamiya, and I. Karube, “Immuno-FET sensor,” J. Mol. Catalysis, vol. 53, no. 3, pp. 285–292, Aug. 1989. [5] A. Offenhusser and W. Knoll, “Cell-transistor hybrid systems and their potential applications,” Trends Biotechnol., vol. 19, no. 2, pp. 62–66, Feb. 2001. [6] Y. Chen, Y.-L. L. Elling, and S. Choong Chong, “A fast, sensitive and label free electrochemical DNA sensor,” J. Phys., vol. 34, pp. 204–209, 2006. [7] M. Schienle, C. Paulus, A. Frey, F. Hofmann, B. Holzapfl, P. SchindlerBauer, and R. Thewes, “A fully electronic DNA sensor with 128 positions and in-pixel A/D conversion,” IEEE J. Solid-State Circuits, vol. 39, no. 12, pp. 2438–2445, Dec. 2004. [8] N. Gemma, S. O’uchi, H. Funaki, J. Okada, and S. Hongo, “CMOS integrated DNA chip for quantitative DNA analysis,” in Solid-State Circuits Conf., 2006, ISSCC 2006. Digest of Technical Papers. IEEE International, 2006, pp. 2288–2297. [9] M. Barbaro, A. Bonfiglio, and L. Raffo, “A charge-modulated FET for detection of biomolecular processes: Conception, modeling and simulation,” IEEE Trans. Electron Devices, vol. 53, no. 1, pp. 158–166, Jan. 2006. [10] A. Caboni, E. Orgiu, E. Scavetta, M. Barbaro, and A. Bonfiglio, “Organicbased sensor for chemical detection in aqueous solution,” Appl. Phys. Lett., vol. 95, no. 12, pp. 123 304–123 306, Sep. 2009. [Online]. Available: http://dx.medra.org/10.1063/1.3232252. [11] M. Barbaro, A. Bonfiglio, L. Raffo, A. Alessandrini, P. Facci, and I. Barák, “Fully electronic DNA hybridization detection by a standard CMOS biochip,” Sens. Actuators B, Chem., vol. 118, no. 1/2, pp. 41–46, Oct. 2006. [12] S. Shao and Y. Wang, “An ultrasensitive field-effect charge sensor for label-free biomolecule detection,” in Proc. IEEE Conf. Lasers ElectroOpt., Aug. 2009, pp. 1–2. [13] P. E. Lobert, D. Bourgeois, R. Pampin, A. Akheyar, L. M. Hagelsieb, D. Flandre, and J. Remacle, “Immobilization of DNA on CMOS compatible materials,” Sens. Actuators B, Chem., vol. 92, no. 1/2, pp. 90–97, Jul. 2003. [14] M. Barbaro, A. Bonfiglio, L. Raffo, A. Alessandrini, P. Facci, and I. Barák, “A CMOS, fully integrated sensor for electronic detection of DNA hybridization,” IEEE Electron Device Lett., vol. 27, no. 7, pp. 595– 597, Jul. 2006. [15] S. Behera, S. Sampath, and C. Retna Raj, “Electrochemical functionalization of a gold electrode with redox-active self-assembled monolayer for electroanalytical application,” J. Phys. Chem. C, vol. 112, no. 10, pp. 3734–3740, 2008. [16] K. E. La Flamme, K. C. Popat, L. Leoni, E. Markiewicz, T. J. La Tempa, B. B. Roman, C. A. Grimes, and T. A. Desai, “Biocompatibility of nanoporous alumina membranes for immunoisolation,” Biomaterials, vol. 28, no. 16, pp. 2638–2645, Jun. 2007. [17] V. P. Parkhutik, J. M. Albella, Y. E. Makushok, I. Montero, J. M. Martinez-Duart, and V. I. Shershulskii, “Study of aluminium anodization in sulphuric and chromic acid solutions,” Electrochim. Acta, vol. 35, no. 6, pp. 955–960, Jun. 1990.

Stefano Lai received the M.S. degree in electronic engineering from the University of Cagliari, Cagliari, Italy, in 2010, where he is currently working toward the Ph.D. degree in the Department of Electrical and Electronic Engineering.

Alessandra Caboni received the Laurea degree from the University of Cagliari, Cagliari, Italy, in 2005 and the Ph.D. degree from the University of Genova, Italy, in 2009. She is currently with the University of Cagliari.

Daniela Loi received the M.Sc. degree from the University of Cagliari, Cagliari, Italy, in 2006 and the Ph.D. degree from the University of Genoa, Genoa, Italy, in 2010. Recently, she joined the R&D Department of RGMD Spa, Genoa.

Massimo Barbaro (M’03) received the M.Sc. and Ph.D. degrees in electronic engineering from the University of Cagliari, Cagliari, Italy, in 1997 and 2001, respectively. He is currently an Assistant Professor with the University of Cagliari.