IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 00, NO. 00, APRIL 2015

1

A Dual-mode Large-arrayed CMOS ISFET Sensor for Accurate and High-throughput pH Sensing in Biomedical Diagnosis Xiwei Huang, Member, IEEE, Hao Yu*, Senior Member, IEEE, Xu Liu, Member, IEEE, Yu Jiang, Mei Yan, and Dongping Wu Abstract—Goal: The existing ISFET based DNA sequencing detects hydrogen ions released during polymerization of DNA strands on microbeads, which are scattered into microwell array above the ISFET sensor with unknown distribution. However, false pH detection happens at empty microwells due to cross-talk from neighbouring microbeads. In this paper, a dual-mode CMOS ISFET sensor is proposed to have accurate pH detection towards DNA sequencing. Methods: Dual-mode sensing, optical and chemical modes, is realized by integrating CMOS image sensor (CIS) with ISFET pH sensor, and is fabricated in standard 0.18μm CIS process. With accurate determination of microbead physical locations with CIS pixel by contact imaging, the dual-mode sensor can correlate local pH for one DNA slice at one location-determined microbead, which can result in improved pH detection accuracy. Moreover, towards high-throughput DNA sequencing, a correlated double sampling (CDS) readout that supports large array for both modes is deployed to reduce pixel-to-pixel non-uniformity such as threshold voltage mismatch. Results: The proposed CMOS dual-mode sensor is experimentally examined to show a well correlated pH map and optical image for microbeads with a pH sensitivity of 26.2mV/pH, a fixed pattern noise (FPN) reduction from 4% to 0.3%, and a readout speed of 1200 frames/second (fps). Conclusion: A dual-mode CMOS ISFET sensor with suppressed FPN for accurate large-arrayed pH sensing is proposed and demonstrated with state-of-the-art measured results towards accurate and high-throughput DNA sequencing. Significance: The developed dual-mode CMOS ISFET sensor has great potential for future personal genome diagnostics with high accuracy and low cost. Index Terms—Correlated double sampling, CMOS image sensor, contact imaging, DNA sequencing, ISFET, pH detection.

Manuscript received November 10, 2014; revised March 07, 2015; accepted March 31, 2015. A preliminary version of this work has been reported in IEEE Symposium on VLSI Circuits, Honolulu, HI, USA, June 2014. This work was supported by the Singapore National Research Foundation Proof-of-Concept Grant NRF2011NRF-POCO01-050. Asterisk indicates corresponding author. X. Huang is with the School of Electronics and Information, Hangzhou Dianzi University, Hangzhou 310018, P. R. China, and also with the School of Electrical and Electronic Engineering, Nanyang Technological University, 639798, Singapore. *H. Yu, X. Liu, Y. Jiang, and M. Yan are with the School of Electrical and Electronic Engineering, Nanyang Technological University, 639798, Singapore (e-mail: [email protected]). D. Wu is with the State Key Laboratory of ASIC and System, School of Microelectronics, Fudan University, Shanghai 200433, P. R. China. Digital Object Identifier 10.1109/TBME.201X.XXXXXXX Copyright (c) 2014 IEEE. Personal use of this material is permitted. However, permission to use this material for any other purposes must be obtained from the IEEE by sending an email to [email protected].

I. INTRODUCTION

T

HE traditional ion-sensitive field effect transistor (ISFET) introduced by Bergveld [1] was essentially a metal-oxide semiconductor FET (MOSFET) without poly-silicon gate as shown in Fig. 1(a). Its fabrication was, however, expensive and not compatible with standard CMOS process. As shown in Fig. 1(b), Bausells first integrated ISFET in standard CMOS process with a poly-silicon gate connected to the top metal and oxi-nitride passivation layer [2], making ISFET with low-cost and mass-manufacturable. As a result, the CMOS ISFET can lead to on-chip integration with high-speed and low-noise CMOS readout circuit for large chemical sensor array. These advantages enable the ISFET applications to evolve over years from neuronal sensing to personalized biomedical diagnosis such as DNA sequencing [3-7], where the features of high-throughput, low-cost, and miniaturization are required [8, 9]. DNA sequencing has profound impact on life technologies such as personal genome study, health care, drug development [10]. Towards higher-throughput and lower-cost sequencing, over the decades, there has been a remarkable development of the Next Generation DNA Sequencing (NGS) instruments [11, 12], such as electrical sequencer called Personal Genome Machine (PGM) [5]. The PGM relies on a large-arrayed ISFET sensor chip to detect hydrogen ions (H+) released by DNA polymerase [13]. The DNA slices to be sequenced are initially prepared to attach onto microbeads, which are then distributed into microwell reaction chambers, each of which corresponds to one ISFET pixel below. As such, the detected local pH change can be used for the DNA sequencing analysis. However, as for the ISFET based DNA sequencing, there is significant inaccuracy existed due to two reasons as illustrated in Fig. 1(c). Firstly, as the DNA-templated microbeads are scattered into microwell array by centrifuge spinning, the distribution of microbeads into microwell array is not known [14]. Thus, the measured pH response has no correlation with the physical locations that contain microbeads. If there is no microbead in the microwell, due to crosstalk from neighboring microbeads in the solution, it will lead to false pH value reported. Secondly, to improve sequencing throughput, one needs large-arrayed ISFET pixels per sensor chip at advanced CMOS process nodes [5]. However, larger process variation occurs in the attributes of transistors such as oxide thickness, channel length and width [15]. As such, the pixel-to-pixel threshold voltage VT mismatch, or fixed pattern noise (FPN)

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 00, NO. 00, APRIL 2015

[16], is also increased, which can significantly degrade pH detection accuracy as the ISFET sensor is to detect the change of the threshold voltage VT. Reference Electrode

Reference Electrode

electrolyte Si3N4 Passivation Layer

electrolyte

S

D

Floating Gate Electrode S

Gate Oxide

n+

(a)

n+

n+ P-substrate

D Poly Gate Gate Oxide

(b)

n+ P-substrate

2

microbeads are in direct contact with the sensor surface, the imaging of the microbead distribution can be detected based on the contact imaging principle without lens [22-25]. As such, an accurate pH-image correlation map can be generated to prune the false pH values due to crosstalk for empty microwells. Moreover, one correlated-double-sampling (CDS) readout is developed for both image and pH modes to reduce the pixel-to-pixel VT mismatch, i.e., FPN. As such, the proposed dual-mode sensor can significantly enhance the accuracy of ISFET based DNA sequencing. The remainder of this paper is organized as follows. In Section II, the brief background of ISFET and contact imaging are introduced. Next, the dual-mode CMOS ISFET sensor is presented in Section III. After that, the CDS readout with FPN reduction is discussed in Section IV. The experimental results are discussed in Section V with conclusions in Section VI. II. BACKGROUND

Fig. 1. (a) Traditional ISFET fabricated by special process. (b) ISFET fabricated by standard CMOS process. (c) Existing large ISFET array for DNA sequencing, which shows the problem of sequencing accuracy due to pH crosstalk and VT variation.

To tackle the first problem, microscope imaging of the 2D microbead distribution can be applied, which still requires bulky optics and tedious calibration. Furthermore, due to the small field-of-view (FOV) when zooming in for micrometer-scale spatial resolution, the microscope needs to physically scan the whole array, which would be slow and could lead to more overhead and detection error. For the second problem, process level UV exposure is used in [17] to remove the trapped charge accumulated on the ISFET gates, but it requires an external calibration source with long time to converge. Some other circuit level techniques have also been proposed. In [18], a programmable-gate (PG) ISFET is proposed to program the ISFET operating point by applying voltage bias to the capacitively coupled floating gate and counteract the effect of trapped charge and drift. But it also requires other auxiliary circuits. Based on PG-ISFET, back-end digital auto-calibration technique is developed in [19] to eliminate mismatch. Their main limitation is not applicable for large-arrayed design towards high-throughput DNA sequencing. A reference-electrode based differential readout method is developed in [20], which makes differential measurement of a reference FET (REFET) and ISFET to reduce the noise. However, because REFET and ISFET are not the same device, the noise may not be correlated to cancel. Towards accurate and high-throughput DNA sequencing, we have demonstrated one 64×64 arrayed dual-mode CMOS ISFET sensor in this paper with preliminary results reported in [21]. Firstly, both optical and pH sensing are performed with a dual-mode sensor pixel that integrates the ISFET with CMOS image sensor (CIS) pixel in standard CIS process. Since the

A. ISFET based pH Sensing The ISFET characteristics are affected by the ionic activity of the electrolyte. In conventional ISFET that has no poly-silicon gate, the immediate pH response results from the surface reaction between electrolyte and ion sensitive membrane on gate oxide. For ISFET fabricated in the standard CMOS process, the pH response is attributed to the surface reaction between electrolyte and the intrinsic passivation layer (Si3N4) as ion sensitivity membrane, which is all the way connected through the metal layers to the poly-silicon gate as shown in Fig. 2. The pH-dependent surface charge causes the changes in ISFET transfer characteristics and modulates its threshold voltage VT.

VREF

Reference Electrode +

+ +

+

+

Electrolyte +

Passivation +

CGouy

Top Metal Diffusion

Poly-Gate

D

S

B

VChem = f(pH)

CHelm Solution Semiconductor CPass CPd CPsub

Well

CPb CPs MOS

Fig. 2. ISFET device model in standard CMOS process.

As shown in Fig. 2, this can be modeled using the site-dissociation model and the Gouy-Chapman-Stern double layer model [26, 27] by VT(ISFET) = VT(MOS) + Vchem,

(1a)

Vchem = γ + 2.3×α·UT· pH

(1b)

and

where Vchem is the chemical voltage between the reference electrode and passivation layer, the only pH dependent item; γ

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 00, NO. 00, APRIL 2015

is a grouping factor of all pH independent terms; UT is the thermal voltage; and 0<α<1 is a sensitivity parameter, whose maximum value gives the theoretical Nernstian sensitivity limit of 59.2 mV/pH at 25°C. What is more, CGouy, CHelm, and CPass are the capacitances of the Gouy–Chapman or diffuse layer, the Helmholtz layer, and the passivation layer, respectively [27]. Besides, the effects of nonidealities determined by different physical design geometries can also be considered. For example, we have parasitic capacitances such as those between passivation to source, CPs; to drain, CPd; to body, CPb; and to substrate, CPsub [28]. As such, the threshold voltage of the ISFET is linearly proportional to pH. B. ISFET based DNA Sequencing As the detection of pH value change can be correlated for DNA sequencing, the CMOS ISFET based sensor has been deployed in PGM [5]. To prepare the genomic sample, the DNA chain is firstly fragmented into slices and ligated to specific adapters, which are then linked to and clonally amplified by emulsion PCR on the surface of microbeads. Next, the templated microbeads are loaded into a microwell array that is fabricated on top of the ISFET chip. The sequencing is primed from a specific location in the adapter sequence. Note that each microbead needs to be loaded into an individual microwell. During sequencing, all four nucleotides (dATP, dGTP, dCTP, and dTTP) are provided sequentially through a microfluidic system to react with the template base of the DNA chain. When they are complementary, the released proton (H+) generates a pH change in the solution of the microwell that is proportional to the number of nucleotides incorporated. As such, the measured pH change at a microbead location indicates the relevant DNA sequence of ATCG [5]. When the pH changes for DNA slices are detected at millions of spatially localized microbeads by a large-arrayed ISFET sensor, a high-throughput DNA sequencing can be thereby realized in a non-optical fashion. However, the main challenge is to improve the sequencing accuracy that can remove false pH reporting as well as non-uniformity. C. Contact Imaging In this paper, in addition to the pH sensing, we will introduce optical sensing for the CMOS ISFET such that a dual-mode sensor can be developed with removal of false pH reporting. Note that conventional optical microscope imaging systems require intermediate bulky lens for magnification, which usually constrains the size, weight, and cost with the difficulty of miniaturization. One promising solution is the use of contact imaging, which directly couples the image sensor array with the sample of interest in small proximity (or contact), as shown in Fig. 3. As such, the sample image can be captured by directly projecting light through it with a detected shadow [23-25]. Contact imaging is kind of near-field sensing without optic lens [24]. As such, contact imaging systems have different geometrical constraints over spatial resolution compared with lens based imaging. In conventional optical imaging systems, the image resolution is determined by the number of pixels in the photo detect array as the scene is entirely projected to the sensor array by optics. By increasing the number of pixels, the spatial resolution for the conventional imaging system can be increased. Differently in the contact imaging, as the image is

3

directly projected from the object to the image sensor array, the resolution is mainly determined by the pixel dimension as well as proximity distance. Thus, the contact imaging is quite suitable for miniaturized biomedical applications to detect objects such as microbeads [25] used in DNA sequencing. Thereby, if one can leverage a dual-mode ISFET sensor with both pH sensing to detect H+ at one microbead and also contact imaging to detect the existence of microbead, the false pH reporting problem of the existing ISFET sensor can be resolved during the DNA sequencing.

Fig. 3. Contact imaging principle: with light source illuminated from above, the contact shadow images of microbeads can be captured by the sensor underneath.

III. DUAL-MODE CMOS ISFET SENSOR As discussed in the introduction, the H+ diffusion of neighboring nucleotide incorporation processes can introduce crosstalk with false pH reporting, which is the fundamental limitation of single-mode CMOS ISFET sensor due to the unknown locations of microbeads [14]. In this paper, we show that with the use of additional contact imaging, one can develop a dual-mode CMOS ISFET sensor with accurately reported pH value for each microbead. A. Dual-mode Sensor Architecture In this paper, a dual-mode CMOS ISFET sensor is developed in standard CMOS image sensor (CIS) process. The cross-sectional view of the proposed dual-mode pixel is shown in Fig. 4(b). Note that n+ and p+ guard rings are placed around the pixel array to minimize the noise generated from the peripheral circuitry. Each pixel is in dual-mode to correlate the local pH value to the existence of one microbead detected by the contact imaging. Therefore, the false pH value reporting problem can be pruned. Moreover, for a large-arrayed ISFET sensor array with the minimum non-uniformity of pixel-to-pixel ISFET threshold voltage mismatch, the CDS readout is further deployed. As the architecture diagram shown in Fig. 4(a), the dual-mode CMOS ISFET sensor contains a 64×64 CIS-ISFET pixel array. All pixels are read out row by row. The row decoder and row driver sequentially address each row, and 64-column of pixel outputs are scanned out sequentially controlled by column decoder. The CDS is implemented with the column sample-and-hold (S/H) circuitry and the global switched-capacitor operational amplifier. After amplifying the difference of reference and signal voltages, a 12-bit pipelined A/D converter converts the results to digitally coded outputs that correspond to pH values.

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 00, NO. 00, APRIL 2015

The SREG is a static register to set the operation mode of the sensor. The IDAC provides the needed on-chip bias current. The digital timing generation is realized off-chip using an FPGA to control the system operation and coordinate all parts of the circuits. Row Decoder/Driver

CIS-ISFET Pixel Array 64×64

SREG

Column Sample/Hold Column Decoder/Driver

IDAC

Global 12-bit AMP Pipeline ADC 2×

S/H sub-ADC

MDAC

1.5bits

S/H

Stage1 1.5bits

Stage2 1.5bits

Stage10 1.5bits

Stage11 2bits

Digital Correction

12bits

(a)

(b)

Fig. 4. (a) Architecture of the dual-mode sensor. (b) Cross-sectional view of the dual-mode pixel layout with microbead contact imaging and ion sensing for DNA sequencing.

B. Dual-mode Pixel Structure Fig. 5 compares the (a) four transistors 4T-CIS pixel and (c) ISFET pixel with the proposed (b) dual-mode pixel. As shown in Fig. 5(b), each dual-mode pixel contains a 4T-CIS pixel to sense the shadow image of microbead by the contact imaging [22-25]. Meanwhile, the source follower (SF) can work as ISFET to detect pH value at one microbead. VAAPIX

VAAPIX

M1

TX M6

VAAPIX

M1

TX RST FD

M6

SF

PD VREF

M3 ROW

PIXOUT

M1

VREF Cpass

M3 ROW

M2

ROW

(a) Capture Contact Image 64 Rows 64 Columns

ROW

VLN_CASC M4

VLN_CASC M4

VLN M5

VLN M5

VLN M5

(b) Dual-mode Pixel

(c) After Loading Solution 64 Rows 64 Columns

RST TX ISFS ISFR CLAMP

PIXOUT

Pixel to Column Sampling

Global Amp & Digitalizing

CDS Timing for Optical Mode

VLN_CASC M4

(a) 4T-CIS Pixel

(b) Before Loading Solution 64 Rows 64 Columns

DOUT

M3 PIXOUT

C. Dual-mode Operation

CB COL ΦADC

SF

SF M2

Cpass

M2 PD

FD

row-select transistor M3 is used to isolate different pixel outputs, and is enabled only when the row is selected for readout. The cascade current source (M4 and M5) provides biasing current and is shared by the whole column for better current matching. In the chemical mode, the poly-gate of SF (M2) is all-the-way connected to the top metal and Si3N4 passivation layer, acting as ion-sensitive membrane of ISFET. Since the change of ion (H+) concentration (or pH) can cause the proportional VT shift of the SF, the corresponding voltage signal is correlated to the pH value that is read out through the source of SF. Considering that VT variation exists in the ISFET transistor, i.e., the SF transistor M2 in Fig. 5, it will show as an offset added at the source follower output PIXOUT, i.e., VPIXOUT=α·(VFD–VT), where α is the gain of the source follower, VFD and VPIXOUT are the input and output voltage of the source follower. Moreover, note that although ISFET pixel has a switch to the floating gate, the TX leakage has been reduced through process optimization from the CIS aspect. As the cross-sectional pixel layout shown in Fig. 4(b), a completely depleted pinned photodiode pixel is used, which consists of a pinned diode (p+-n+-p) to reduce the surface-defect noise due to dark current. The depletion layer of a pinned photodiode stretches almost to the Si–SiO2 interface, which is perfectly shielded by the p+ layer that keeps the interface fully filled with holes, making the leakage extremely low [29,30].

SHR SHS Φ1 Φ2

RST

RST

4

(c) ISFET Pixel

Fig. 5. Dual-mode pixel schematic in comparison with 4T-CIS pixel and ISFET pixel.

In the optical mode, photodiode PD first collects photons and converts them to proportional electrons, whose drifting generates the photo current. As the intrinsic junction capacitor of PD can store the generated charges, after a certain integration period, the intensity of incident light carried by the amount of charges is translated to a voltage signal. The charges can be transferred to floating diffusion (FD) by turning on ‘TX’ switch of M6. As such, the voltage signal indicating the shadow image of microbead is detected through contact imaging. Then, the corresponding voltage signal for the optical image is buffered by SF (M2) and read out to PIXOUT node through its source under the control of 'ROW' select-signal of M3. Since there are multiple rows of pixels that share the same PIXOUT line, the

Pixel to Column RST Sampling

Global Amp & Digitalizing

Pixel to Column SIG Sampling

Global Amp & Digitalizing

CDS Timing for Chemical Mode

Fig. 6. Dual-mode readout timing for both (a) optical mode and (b) chemical mode.

Based on the dual-mode pixel structure, the timing of dual-mode CIS-ISFET pixel sensing control is operated as follows corresponding to the readout timings shown in Fig. 6 with following steps. 1) When reaction carrier microbeads are initially distributed into the sensor pixel array, the readout timing is set to optical mode shown in Fig. 6(a) as a normal 4T-CIS pixel. The shadow images of microbeads can be captured by the contact imaging. After that, the existence of microbead at each pixel can be determined with an address generated. 2) Then the optical mode changes to chemical mode. Before loading ATCG solution, the reference reset-signal for the whole pixel array is read out using the timing in Fig. 6(b). 3) After loading ATCG solution sequentially, the pH readout timing changes to Fig. 6(c) to obtain the signal of the pixel array with actual pH value at individual microbead. As such, one can obtain the accurate correlation between the

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 00, NO. 00, APRIL 2015

measured pH data and the distribution of microbeads. The false pH data at empty microwells can be thereby eliminated by the dual-mode operation. IV. CDS READOUT As ISFET sensor senses the VT, the pixel-to-pixel VT mismatch in large-arrayed sensor can introduce significant error. Note that CDS is commonly used to reduce the VT mismatch and improve the signal-to-noise ratio during readout [16]. In this paper, the CDS for the VT mismatch cancellation and FPN reduction is developed in the dual-mode CMOS ISFET sensor array. The CDS readout timing is shown in Fig. 6 for readout. The rolling shutter reset phase with PD reset through TX pulse is not illustrated for simplicity of illustration. Note that the developed CDS is applied to suppress pixel-to-pixel VT mismatch by using each pixel itself as reference, which is intrinsically better than the differential measurement using another REFET device in [20]. In addition as shown in Fig. 7, the whole CDS readout-chain schematic includes column sample/hold, global amplifier, and pipelined analog-to-digital converter (ADC). VCM (1pF,0.5pF,0.25pF) Φ1

VAAPIX RST TX FD

CPass PD

VREF

SF

CFS SHS VPOS

ISFET ROW

VLN_CASC

CSS(1pF)

CLAMP

ISFR

VCM

CB ISFS

SHR VPOR

COL VINP

CLAMP

Φ1P

Global Opamp

CSR(1pF) COL VINN

VLN

Dual-Mode Pixel

Φ1P

VCM

CFR

Column S/H

(1pF,0.5pF,0.25pF)

VCM

ΦADC

Φ2 VOUTN Φ1 12-bit DOUT Pipelined ADC

Φ2

VOUTP

Φ1

VREFN VREFP

VCM

Fig. 7. CDS readout schematic for dual-mode sensor.

A. CDS Readout for CIS We first discuss the CDS for CIS in optical mode. As the timing shown in Fig. 6(a), during the pixel to column readout period, 'CLAMP' switch is on so that the top plate of sampling capacitors CSS and CSR are clamped to VCM. The charges on the pixel output line are sampled to CSR when 'SHR' switch is on and to CSS when 'SHS' switch is on. These correspond to the pixel reset level voltage VRST and signal level voltage VSIG at the source follower input node. Then during the column to amplifier readout period, the 'CLAMP' switch is off. The crow-bar switch 'CB' is off during the amplifier reset phase and is on during the charge amplifying phase. The value of sample and hold (S/H) capacitor is determined by the balance between the KTC noise and speed. A 1 pF poly capacitor is chosen for both CSS and CSR, i.e., CSS = CSR = CS = 1 pF. The output of the column sample capacitors are successively controlled by the column select signals 'COL'. Following the column S/H is the global switched-capacitor amplifier that consists of one non-overlapping clock generator, a fully differential cascode amplifier with switched-capacitor common-mode feedback (CMFB), and several poly capacitors CFS and CFR for programmable gain control. We also have CFR = CFR = CF. The non-overlapping clock generator generates a pair of

5

non-overlapping clock signals: 'Φ1' and 'Φ2', and 'Φ1P' whose falling edge is slightly earlier than 'Φ1' to reduce the possible charge injection and clock feed-through. 'Φ1' and 'Φ2' work under 'CB' to amplify and read out signals of each column. During the reset phase Φ1, both inputs and outputs of the amplifier as well as the feedback capacitor CF are reset to VCM, the common mode voltage. During the amplify phase Φ2, the bottom plates of the sample and hold capacitors are shorted by turning on 'CB' for the currently selected readout column; and the feedback capacitor CF are connected to the amplifier output. Thus, charges are essentially moved from the column sample and hold capacitor CS to the feedback capacitor CF. As the two input nodes of the differential amplifier connect with the pixel output reset level VPOR and pixel output signal level VPOS, and VPOR=α·(VRST–VT), VPOS=α·(VSIG–VT), thus only the difference between them is amplified and output, i.e., VOUT= VOUTP–VOUTN= α·(CS/CF)·[(VRST –VT)–(VSIG–VT)] (2) = α·(CS/CF)· (VRST–VSIG) where α is the gain of the source follower. As such, the VOUT removes the dependence on VT for CIS in the optical mode. B. CDS Readout for ISFET We add switches 'ISFR'/'ISFS' as shown in Fig. 7 to realize the CDS for ISFET in pH mode. Other readout circuits remain the same. The corresponding timing diagram is shown in Fig. 6(b)-(c). The CDS readout for ISFET is performed as follows. As shown in Fig. 6(b), before loading solution with microbeads, 'RST' is turned on, and the reset voltage VRST is stored at sampling capacitor CSR by turning on 'SHR'. Meanwhile, 'ISFR' is turned on to force VPOS =VCM. As such, the output of the amplifier is VOUT1=VOUTP–VOUTN=α·(CS/CF)· (VRST–VT–VCM).

(3)

Afterwards the reset voltage level for the whole array is read out and digitized by the 12-bit pipelined ADC at the next stage and saved by the external storage. As shown in Fig. 6(c), after loading solution with microbeads, 'ISFS' is turned on to force VPOR=VCM. The amplifier output now is VOUT2=VOUTP–VOUTN=α·(CS/CF)·(VREF–VCM–VT–dV), (4) where dV is the threshold voltage change caused by the chemical reaction between the ion and the passivation layer; and VREF is the voltage of the reference electrode. This output is also converted by the ADC and readout to the external storage for further digital processing. As there is an equivalent passivation capacitor Cpass connected to the ISFET gate, the DC level of electrode VREF is thereby not equal to the FD voltage. However, note that a reference electrode in the analyte solution determines the electric potential of the bulk of the analyte solution, and always provides a fixed electrical reference. Therefore, the difference of the VREF and the FD node can be considered as the dV. As such, as long as the VT is subtracted, the dependence on VT for the ISFET in the chemical mode can be removed. As a result, we subtract the two outputs and obtain the

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 00, NO. 00, APRIL 2015

difference by VOUT1–VOUT2=α·(CS/CF)·(VRST–VREF+dV),

(5)

which removes the dependence on VT for ISFET in the chemical mode. Hence, CDS readout circuit measures each individual ISFET pixel output voltage level before the chemical reaction, and use this as reference to calibrate the DNA sequencing measurement data for higher accuracy. C. Global Amplifier As for the global amplifier, the feedback capacitors CF are adjustable among 1 pF, 0.5 pF and 0.25 pF such that the gain CS/CF can be selected among 1X, 2X, and 4X under different input signal levels. Thus, the input referred noise of the gain amplifier has a reduction factor of √𝐺, where G is the close loop gain of the column readout amplifier. The sensitivity or dynamic range can then be improved. The amplifier utilizes a telescopic structure. AC simulation results show an open-loop gain of 68 dB and bandwidth of 628 MHz. The high gain-bandwidth product (GBW) enables high-speed readout with 10 MHz column-wise readout speed. D. Pipelined ADC The sensed signal by ISFET-sensor array is digitized by 12-bit pipelined ADC before the final output. The ADC consists of sample/hold input stages, ten serially connected 1.5-bit pipeline stages, and one 2-bit flash stage. The digital correction block creates a 12-bit output code by redundant signed digit (RSD). The 1.5-bit per-stage is chosen because of its immunity to the offsets. A telescopic operational amplifier with gain-boosting is chosen for high dc gain, high GBW and fast settling time. The maximum differential nonlinearity (DNL) is 0.44 LSB and the maximum integral nonlinearity (INL) is 0.61 LSB. The effective number of bits (ENOB) is 11.4 bits, and the signal-to-noise and distortion ratio (SNDR) is 70.35 dB. As such, the whole row-readout time is 13 μs, including the pixel sampling, amplification and digitization. Therefore, for a 64×64 array, the whole readout time for 64 rows is 64×13 μs = 0.832 ms with a readout frame rate about 1/0.832 ms = 1200 fps. Fast readout speed can enable us to capture a chemical image in a shorter time. For DNA sequencing, sampling the signal at high frequency relative to the time of the nucleotide incorporation signal allows signal averaging to improve the SNR.

6

the threshold voltage VT of the ISFET. To evaluate the effect of negative charge, i.e., donor, and positive charge, i.e., acceptor, we choose VDS=1V with only donor or accepter change in density. The addition of donors and acceptor to semiconductor is similar to the process of adding acids or bases to pure water and altering the balance between H+ an OH–.

Fig. 8. Cross-sectional structure of the ISFET as modeled in Sentaurus TCAD. The electron concentration during ISFET operation is indicated by coloring.

The cross-sectional structure of the ISFET modeled is shown in Fig. 8. The electron concentration is indicated by coloring corresponding to the bar graph on the right. As the results shown in Fig. 9 (a), when we increase the concentration of negative charge donor from 1016/cm-3 to 1040/cm-3 with acceptors=0, the VT has a corresponding linear reduction from 1.11 V to 0.06 V. As in Fig. 9 (b), when increasing the concentration of positive charge acceptor from 1016/cm-3 to 1036/cm-3 with donor=0, the VT has a corresponding linear increase from 1.13 V to 1.74 V. A natural logarithm scale for charge concentrations is used. As such, although the electrolyte is not directly modeled, the effect of changing the surface charge will cause the changes in ISFET transfer characteristics and linearly modulates the ISFET threshold voltage VT, which is the basic principle of ISFET based pH sensing.

V. RESULTS A. ISFET Device Modeling To evaluate the dual-mode ISFET device fabricated by standard CMOS image sensor process, we model an N-type ISFET with six metal layers using Synopsys® Sentaurus TCAD (Technology Computer-Aided Design), which is a suite of commercial TCAD tools that simulates the fabrication, operation and reliability of semiconductor devices. As the electrolyte cannot be directly modeled, in the simulation, we change the concentration of the trapped charge in poly-silicon gate area to try to simulate the charge change in poly-silicon caused by the ion reaction on the surface of the passivation layer. The charge change in the poly will result in the change of

(a)

(b)

Fig. 9. ISFET device simulation results showing the threshold voltage VT change.

B. Testing System Setup The proposed dual-mode ISFET sensor is fabricated in standard TSMC 0.18μm CIS process. After fabrication, the chip is packaged in a 100-pin Pin Grid Array (PGA) package with a size of 33.5 mm × 33.5 mm. As the experimental processes need to be conducted in aqueous environments, proper encapsulation of the sensor chip is necessary to protect the circuits. Thus, we use epoxy to encapsulate the whole chip

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 00, NO. 00, APRIL 2015

with the sensing pixel array area open only, as shown in Fig. 10 (b) and (c). Meanwhile, the bonding wires and bonding pads are also covered by epoxy. To retain aqueous samples on the top of sensor chip, in addition, a 3D-printed plastic reservoir that just fits the PGA package is mounted on the package with epoxy to fill the gap at all four sides. The plastic reservoir is also designed to be able to fix the Ag/AgCl reference electrode. The package chip is then mounted on a specially designed printed-circuit-board (PCB) through a 100-pin PGA socket. The PCB, which is further connected with a Xilinx Virtex-6 XC6VLX240T FPGA demo board [31], is designed to provide power supply and digital timing control signals to the sensor chip. We measured the electrochemical characteristics of the chip under the control of a MATLAB-based (Mathworks, Natick, MA) Graphical User Interface (GUI). The chip micrograph with architecture and testing system is shown in Fig. 10. The design specifications are summarized in Table I.

Row Decooder

(a) Testing PCB Setup with FPGA

7

microbead compared with 10μm pixel size is selected such that the contrast of shadow imaging can be better. With the contact shadow imaging, the image size of microbead takes up about a 5×5 pixel array area. Due to the diffraction effect, the center pixels show darker intensity and the pixels near the boundary show lighter intensity. For the proof-of-concept verification, the microbeads are first diluted and prepared in acid solution as they are ideally suited for protein binding using passive adsorption techniques, and then dropped onto the sensor surface to test the local pH changes. The contact image determines the existence of microbeads and provides their addressed distribution. The exposure time of the contact imaging is 160μs. The pH map is thereby locally associated with microbeads by pruning out those uncorrelated pH data. Due to the diffusion effect, the pH map at microbead locations show a pattern similar to normal distribution.

ISFETCIS Pixel Array

Column S/H Column Decoder (c)

3D Printed Framework Silicone Rubber

(b) Ion-Image Sensor with Liquid-friendly Encapsulation

PGA Package

SREG Bandgap

Open Sensing Pixel Array

Ion-Image Sensor Die

(a) Contact Image AMP

Pipelined ADC

Encapsulated Bonding Wires

(d)

Fig. 10. (a) FPGA based testing system setup. (b) Ion-image sensor with liquid friendly encapsulation with 3D-printed plastic reservoir attached on the chip PGA package. (c) Cross-sectional view of the encapsulated packaging strategy. (d) Micrograph photo of the dual-mode sensor chip. TABLE I SPECIFICATIONS OF DUAL-MODE SENSOR Parameters Process Pixel Type Pixel Size Pixel Optical Sensing Area Pixel Chemical Sensing Area Array Size Die Area ADC ENOB ADC SNDR FPN Frame Rate Total Power Consumption

(b) pH Map

Fig. 11. The correlated maps of distributed microbeads: (a) contact images and (b) pH values.

Specifications Standard TSMC 0.18μm CIS Dual-Mode (Image and Chemical) 10μm×10μm 2

20.1μm (Fill Factor=20.1% ) 2 22.3μm (Fill Factor =22.3% ) 64×64 2.5mm×5mm 11.4 bits 70.35dB 0.3% 1200fps 32mA @ 3.3V

C. Measurement of Correlated Ion-Image Map Firstly, the correlated contact image and pH map of microbeads are shown in Fig. 11. The microbeads of 45μm diameter are used (Product# 07314-5, Polysciences, Warrington, PA). Note that we have not fabricated the microwell array on top of the image sensor die to correspond each microwell with an ISFET pixel. Thus, a relatively larger

D. Characterization of Dual-mode Sensor To characterize the pH sensing capability of the dual-mode sensor, the pH sensitivity is tested and the measurement results are shown in Fig. 12(a). The pH of solution is changed by adding HCL and NaOH. The pH readout sensitivity of ISFET by CIS process is measured as 26.2 mV/pH with amplifier gain=1 and as 103.8 mV/pH with amplifier gain=4 as sown in Fig. 12(a). The device sensitivity at gain=1 is somewhat lower than the commonly observed response of 45–56 mV/pH for Si3N4 [32], this can be due to the low-pressure chemical vapor deposition (LPCVD) technique for Si3N4S at low temperature, which generally cause low-density and porous passivation layer. It can be optimized by the LPCVD at a high temperature or do additional depositions, which are still standard CMOS process [14].

(a)

(b)

Fig. 12. Measurement results: (a) pH sensitivity of dual-mode ISFET sensor, and (b) the comparison with commercial pH meter for bacteria (E. Coli) culture solution with glucose at different time intervals.

The CMOS ISFET sensor chip is also calibrated by testing

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 00, NO. 00, APRIL 2015

the pH change of a bacteria (Escherichia coli) culture solution at different time intervals and comparing with commercial tool. By extracting the sample solution of the bacteria culture for testing at 1 to 6 hour time intervals, the measurement results by the dual-mode sensor can correlate well with one commercial pH meter (Checker, Hanna Instruments, RI, US) in Fig. 12(b). Max=0.71mV

8

large-arrayed sensor, the pH sensing accuracy can be significantly improved. Our future work is to perform the actual DNA sequencing testing using the proposed sensor, which requires fabricating the microwell array on the sensor, and building the microfluidic system for sample loading. We believe that the developed dual-mode CMOS ISFET sensor has great potential for the future personal genome diagnostics.

Mean=0.26mV

REFERENCES [1]

[2]

[3]

Mean=0.09mV

Max=0.46mV

[4]

[5] Fig. 13. Measurement results: spatial FFT of readout voltage variations (a) without CDS and (b) with CDS readout.

The comparison of readout voltage variations with and without CDS is shown in Fig. 13. After performing spatial FFT to the readout voltages with respect to the addresses of the sensor array, the mean and peak variations are reduced by 0.17 mV and 0.25 mV, respectively. The FPN is accordingly reduced from 4% down to 0.3%. Lastly, the comparisons with the state-of-the-art ISFET sensors are summarized in Table II. The proposed dual-mode sensor shows the state-of-the-art results: 10 μm pixel pitch, 64×64 pixel array, fast frame rate of 1200 fps, and readout sensitivity of 103.8 mV/pH in standard CIS process.

[6]

[7]

[8]

[9]

[10] [11] [12]

TABLE I COMPARISON OF STATE-OF-THE-ART ISFET SENSORS [33] [34] [35] [36] [37] 5μm 0.35μm 0.35μm 0.18μm 0.35μm Process NonModified Standard Standard Standard CMOS CMOS CMOS CMOS CMOS 200μm× 12.8μm× 10.2μm× 20μm× 150μm× Pixel Size 200μm 12.8μm 10.2μm 2μm 150μm Array Size 10×10 16×16 64×64 8×8 8×8 Frame Rate 30fps 100fps 6fps Sensitivity (mV/pH)

229

46

20

37

57

Dual-Mode

No

No

No

No

Yes

This Work 0.18μm Standard CMOS 10μm× 10μm 64×64 1200fps 26.2 (gain=1) 103.8 (gain=4) Yes

[13] [14] [15]

[16]

[17]

[18] [19]

VI. CONCLUSIONS In this article, a dual-mode CMOS ISFET sensor with suppressed FPN for accurate large-arrayed pH sensing is proposed and demonstrated with state-of-the-art results, targeted for accurate and high-throughput DNA sequencing. With the dual-mode pixel design to prune false pH value reporting by determining the existence of microbead with contact imaging; and the CDS readout to suppress FPN for

[20]

[21]

[22]

P. Bergveld, “Development of an ion-sensitive solid-state device for neurophysiological measurements,” IEEE Trans. Biomed. Eng., vol. 17, no. 1, pp. 70–71, Jan. 1970. J. Bausells et al., “Ion-sensitive field effect transistors fabricated in a commercial CMOS technology,” Sens. Actuators B Chem., vol. 57, pp. 56–62, Sep. 1999. P. Bergveld, “Thirty years of ISFETOLOGY What happened in the past 30 years and what may happen in the next 30 years,” Sens. Actuators B Chem., vol.88, no.1, pp. 1–20, Jan. 2003. C. Toumazou, and P. Georgiou, “Piet Bergveld – 40 years of ISFET technology: from neuronal sensing to DNA sequencing,” Electron. Lett., vol. 47, no. 26, pp. 7–12, Dec. 2011. J. M. Rothberg et al., “An integrated semiconductor device enabling non-optical genome sequencing,” Nature, vol. 475, pp. 348–352, Jul. 2011. C. Toumazou et al., “Simultaneous DNA amplification and detection using a pH-sensing semiconductor system,” Nat. Methods, vol. 10, pp. 641–646, Jun. 2013. A. Manickam et al. “A fully-electronic charge-based DNA sequencing CMOS biochip,” IEEE Symp. VLSI Circuits Dig. Tech. Papers, Jun. 2012, pp. 126–127. A. Pai et al., “A handheld magnetic sensing platform for antigen and nucleic acid detection,” Analyst, vol. 139, no. 6, pp. 1403–1411, Dec. 2013. K. Venta et al., “Differentiation of short, single-stranded DNA homopolymers in solid-state nanopores,” ACS Nano, vol. 7, no. 5, pp. 4629–4636, May 2013. M. Margulies et al., “Genome sequencing in microfabricated high-density picolitre reactors,” Nature, vol. 437, pp. 376–380, Sep. 2005. E. R. Mardis, “Next generation DNA sequencing methods,” Ann. Rev. Genomics Hum. Genet., vol. 9, pp. 387–402, Jun. 2008. A. Ahmadian, and H. A. Svahn, “Massively parallel sequencing platforms using lab on a chip technologies,” Lab Chip, vol. 11, no. 16, pp. 2653–2655, Apr. 2011. N. Pourmand et al., “Direct electrical detection of DNA synthesis,” Proc. Natl. Acad. Sci. USA, vol. 103, no. 17, pp. 6466–6470, Apr. 2006. J. M. Rothberg et al., “Methods and apparatus for measuring analytes,” U.S. Patent 20100301398, Dec. 02, 2010. B. E. Stine et al., “Analysis and decomposition of spatial variation in integrated circuit processes and devices,” IEEE Trans. Semicond. Manuf., vol. 10, no. 1, pp. 24–41, Feb. 1997. K. Yonemoto, and H. Sumi, “A CMOS image sensor with a simple fixed-pattern-noise-reduction technology and a hole accumulation diode,” IEEE J. Solid-State Circuits, vol. 35, no. 12, pp. 2038–2043, Dec. 2000. M. J. Milgrew, and D. R. S. Cumming, “Matching the transconductance characteristics of CMOS ISFET arrays by removing trapped charge,” IEEE Trans. Electron Devices, vol. 55, no. 4, pp. 1074–1079, Apr. 2008. P. Georgiou, and C. Toumazou, “CMOS-based programmable gate ISFET,” Electron. Lett., vol. 44, no. 22, pp. 1289–1290, Oct, 2008. C. Z. D. Goh et al., “A CMOS-based ISFET chemical imager with autocalibration capability,” IEEE Sensors J., vol. 11, no. 12, pp. 3253–3260, Dec. 2011. P. A. Hammond et al., “Design of a single-chip pH sensor using a conventional 0.6-μm CMOS process,” IEEE Sensors J., vol. 4, no. 6, pp. 706–712, Dec. 2004. X. Huang et al., “A 64x64 1200fps CMOS ion-image sensor with suppressed fixed-pattern-noise for accurate high-throughput DNA sequencing,” IEEE Symp. VLSI Circuits Dig. Tech. Papers, Jun. 2014, pp. 109–110. J. Guo et al., “Portable resistive pulse-activated lens-free cell imaging system,” RSC Adv., vol. 4, pp. 56342–56345, Oct. 2014.

IEEE TRANSACTIONS ON BIOMEDICAL ENGINEERING, VOL. 00, NO. 00, APRIL 2015

[23] X. Huang et al., “A contact-imaging based microfluidic cytometer with machine-learning for single-frame super-resolution processing,” PLoS ONE, vol. 9, no. 8, pp. e104539, Aug. 2014. [24] H. Ji et al., “Contact imaging: simulation and experiment,” IEEE Trans. Circuits Syst. I, Reg. Papers, vol. 54, no. 8, pp. 1698–1710, Aug. 2007. [25] A. Ozcan, and U. Demirci, “Ultra wide-field lens-free monitoring of cells on-chip,” Lab Chip, vol. 8, no. 1, pp. 98–106, Jan. 2008. [26] L. Shepherd, and C. Toumazou, “Weak Inversion ISFETs for ultra-low power biochemical sensing and real-time analysis,” Sens. Actuators B Chem., vol. 107, no. 1, pp. 468-473, May 2005. [27] S. Martinoia, and G. Massobrio, “A behavioral macromodel of the ISFET in SPICE,” Sens. Actuators B Chem., vol. 62, no. 3, pp. 182–189, Mar. 2000. [28] Y. Liu et al., “An extended CMOS ISFET model incorporating the physical design geometry and the effects on performance and offset variation,” IEEE Trans. Electron Devices, vol. 58, no. 12, pp. 4414–4422, Dec. 2011. [29] E. R. Fossum, and D. B. Hondongwa, “A review of the pinned photodiode for CCD and CMOS image sensors,” IEEE J. Electron Devices Soc., vol. 2, no. 3, pp. 33–43, May 2014. [30] A. J. P. Theuwissen, “CMOS image sensors: State-of-the-art,” Solid State Electron., vol. 52, no. 9, pp. 1401–1406, Sep. 2008. [31] Xilinx Virtex-6 XC6VLX240T FPGA (2014, Oct). [On-line] Available: http://www.xilinx.com/products/boards-and-kits/EK-V6-ML605-G.htm [32] D. L. Harame et al., “Ion-sensing devices with silicon nitride and borosilicate glass insulators,” IEEE Trans. on Electron Devices, vol. 34, no. 8, pp. 1700–1707, Aug. 1987. [33] T. Hizawa et al., “Fabrication of a two-dimensional pH image sensor using a charge transfer technique,” Sens Actuators B Chem., vol. 117, no. 2, pp. 509–515, Oct. 2006. [34] M. J. Milgrew et al., “A 16x16 CMOS proton camera array for direct extracellular imaging of hydrogen-Ion activity,” ISSCC Dig. Tech. Papers, Feb. 2008, pp. 590–638. [35] B. Nemeth et al., “High-resolution real-time ion-camera system using a CMOS-based chemical sensor array for proton imaging,” Sens. Actuators B Chem., vol. 171, pp. 747–752, Sep. 2012. [36] W. P. Chan et al., “An integrated ISFETs instrumentation system in standard CMOS technology,” J. Solid State Circuits, vol. 45, no. 9, pp. 1923–1934, Sep. 2010. [37] C. Z. D. Goh et al., “Live demonstration: A CMOS-based lab-on-chip array for combined magnetic manipulation and opto-chemical sensing,” in Proc. IEEE Int. Symp. Circuits Syst., May 2011, pp. 1997–2001.

Xiwei Huang (S'10-M'15) received the B.Eng. degree from Beijing Institute of Technology (BIT), China in 2009, and the Ph.D. degree in electronic engineering from Nanyang Technological University (NTU) in March 2015. He joined Hangzhou Dianzi University as an Assistant Professor in May 2015. His primary research interest is CMOS multimodal sensor for personalized biomedical diagnostics, including CMOS image and ISFET sensor design, microfluidic on-chip imaging system.

Hao Yu (M’06-SM’14) received the B.S. degree from Fudan University, China; and Ph.D degree from the Electrical Engineering Department, University of California, USA. He was a senior research staff at Berkeley Design Automation. Since October 2009, he has been an Assistant Professor at the School of Electrical and Electronic Engineering, Nanyang Technological University, Singapore. His primary research interests are 3D-IC and RF-IC at nano-tera scale. Dr. Yu received Best Paper Award from the ACM TODAES’10, Best Paper Award nominations in DAC’06, ICCAD’06, ASP-DAC’12, Best Student Paper (advisor) Finalist in SiRF’13, RFIC’13, and Inventor Award’08 from semiconductor research cooperation. He is associate editor and technical program committee member for a number of journals and conferences.

9 Xu Liu (M’10) received the B.S. and M.S. degrees in electronics engineering from Huazhong University of Science and Technology (HUST), Wuhan, China, in 2007 and 2009. He received Ph.D. degree in electrical and computer engineering from National University of Singapore (NUS), Singapore, in 2015. He is currently with Nanyang Technological University (NTU), Singapore. His research interests are biomedical IC design, bio-sensors, and mixed-signal IC design.

Yu Jiang received the B.Eng. degree in electronics and information engineering from Anhui University, Anhui, China, and the M.S degree in electronics science and technology from Fudan University, Shanghai, China, in 2011 and 2014, respectively. She is currently a Research Associate at Nanyang Technological University. Her research interest is on biosensor and image sensor design for biomedical applications.

Mei Yan received the Ph.D. degree from State University of New York at Stony Brook in 2004. During her Ph.D. study, she worked with scientists in US Brookhaven National Lab, and developed the world’s first linear Digital CMOS Image Sensor for barcode scanner. Then she worked for Aptina Imaging for 6 years, where she was involved in more than 10 advanced products design. Dr. Yan joined NTU as Research Fellow in Oct. 2010. Her research interest is to build innovated CMOS based biomedical imaging system by integrating image sensor design with last advanced device development such as MEMS, microfluidic and ISFET sensor.

Dongping Wu received the Ph.D. degree in department of microelectronics and information technology from Royal Institute of Technology(KTH) of Sweden in 2004. He is currently a professor in the school of microelectronics, Fudan University, China. His research interest covers nano-scale semiconductor devices, semiconductor sensors, and bioelectronics.