A Time-Resolved CMOS Image Sensor With Draining ... - IEEE Xplore

IEEE TRANSACTIONS ON ELECTRON DEVICES, VOL. 59, NO. 10, OCTOBER 2012

2715

A Time-Resolved CMOS Image Sensor With Draining-Only Modulation Pixels for Fluorescence Lifetime Imaging Zhuo Li, Student Member, IEEE, Shoji Kawahito, Fellow, IEEE, Keita Yasutomi, Member, IEEE, Keiichiro Kagawa, Member, IEEE, Juichiro Ukon, Mamoru Hashimoto, and Hirohiko Niioka

Abstract—This paper presents a time-resolved CMOS image sensor with draining-only modulation (DOM) pixels, for timedomain fluorescence lifetime imaging. In the DOM pixels using a pinned photodiode (PPD) technology, a time-windowed signal charge transfer from a PPD to a pinned storage diode (PSD) is controlled by a draining gate only, without a transfer gate between the two diodes. This structure allows a potential barrierless and trapless charge transfer from the PPD to the PSD. A 256 × 256 pixel time-resolved CMOS imager with 7.5 × 7.5 μm2 DOM pixels has been implemented using 0.18-μm CMOS image sensor process technology with PPD option. The prototype demonstrates high sensitivity for weak signal of less than one electron per light pulse and accurate measurement of fluorescence decay process with subnanosecond time resolution. Index Terms—Barrierless, CMOS image sensor, draining-only modulation (DOM), fluorescence lifetime imaging microscopy (FLIM), linearity, low noise, time-domain lifetime measurement, time resolved.

I. I NTRODUCTION

F

LUORESCENCE imaging is a powerful tool in biology. Fluorescence has two physical quantities: intensity and lifetime of decaying. The fluorescence intensity measurement has been widely used. However, it has difficulty in quantitative measurement because the fluorescence intensity is influenced by many factors, such as fluorophore concentration, degradation of fluorophore, wavelength and intensity of excitation flux, sensitivity of detectors, and transmittance of optical system. To

Manuscript received December 7, 2011; revised June 3, 2012; accepted July 9, 2012. Date of publication August 20, 2012; date of current version September 18, 2012. This work was supported in part by the Ministry of Education, Culture, Sports, Science and Technology under Grant-in-Aid for Scientific Research (A) 22246049. The review of this paper was arranged by Editor J. R. Tower. Z. Li is with the Graduate School of Science and Technology, Shizuoka University, Hamamatsu 432-8011, Japan (e-mail: [email protected]). S. Kawahito, K. Yasutomi, and K. Kagawa are with the Research Institute of Electronics, Shizuoka University, Hamamatsu 432-8011, Japan (e-mail: [email protected]; [email protected]; kagawa@idl. rie.shizuoka.ac.jp). J. Ukon is with the Advanced Research and Development Center, Horiba, Ltd., Kyoto 601-8510, Japan (e-mail: [email protected]). M. Hashimoto and H. Niioka are with the Graduate School of Engineering Science, Osaka University, Osaka 560-8531, Japan (e-mail: mamoru@ me.es.osaka-u.ac.jp; [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.2012.2209179

address these difficulties and to provide additional novel information, fluorescence lifetime measurement has been becoming important technology in biological imaging [1]. Fluorescence lifetime measurement uses intensity decay rate rather than the absolute value of intensity, and therefore, the quantification is not influenced by degradation of fluorescence intensity. The fluorescence lifetime imaging microscopy (FLIM) has a variety of applications. It can be used to quantify physical parameters such as microviscosity and chemical parameters such as pH and ion concentration. It can also be a powerful tool in DNA sequencing. The time-correlated single-photon counting (TCSPC) method [2] is a typical method used in FLIM systems. The time difference between excitation and fluorescence detection is calculated to obtain decay waveform of photon number. To avoid pileup effect in the TCSPC method, excitation frequency is limited under 5% of photon counting rate. This limits the dynamic range of the fluorescence lifetime measurements. A typical TCSPC system uses a photomultiplier tube [3] for detecting fluorescence emission, and an expensive mechanical scanning mirror and optical systems are necessary for 2-D imaging, resulting in a bulky and expensive system. Time-resolved CMOS imagers have been paid much attention to implementing a compact and low-cost FLIM system. Single-photon avalanche diodes (SPADs) are known as devices for time-resolved lifetime measurements [4]. In SPADs, when a p-n junction biased above breakdown voltage absorbs a photon, a large current is generated and then rapidly quenched by a load resistance, causing an electric pulse per photon [5], [6]. A SPAD-based time-resolved imager consists of a SPAD detector array with sensing electronics in each pixel, time-todigital converters (TDCs), digital integrators to intensify the signals, and readout electronics. For high photon counting rate, a large number of TDCs and digital integrators are necessary. These features make the SPAD-based time-resolved imagers complicated and limit the spatial resolution. Yoon et al. [7] proposed a CMOS time-resolved imager using a two-stage charge transfer (TSCT) technique in pinned diode for the measurement of fluorescence lifetime. In the previous structure, using a fully depleted pinned photodiode (PPD), fluorescence lifetime of nanosecond timescales has been measured. Because of the simple structure, a high-spatial-resolution fluorescence lifetime imager can be realized. However, the implemented fluorescent lifetime imager has a problem of the

0018-9383/$31.00 © 2012 IEEE

2716


pixel-to-pixel variation of sensitivity at very low light level. To address this problem, this paper presents a time-resolved CMOS image sensor with a draining-only modulation (DOM) pixel structure. The DOM structure removes the transfer gate between the PPD and the pinned storage diode (PSD). This allows us to realize a barrierless charge transfer between the PPD and PSD, leading to high sensitivity of weak fluorescent signals with high time resolution. Using a monotonic positive lateral electric field, high-speed charge transfer from the PPD to the PSD in the timescales of nanosecond is possible [8], [9]. The time windowing is done by draining the charges with a draining gate (TD) only, which is attached beside the carrier path from the PPD to the PSD. A 256 × 256 pixel time-resolved CMOS image sensor chip has been implemented for the proof of concept of the DOM pixel. The rest of this paper is organized as follows. Section II describes the DOM pixel structure and operation. Section III describes the design of DOM pixel and optimization of the pixel structure using device simulations. Section IV provides the measurement results of linearity and fluorescence lifetime. Finally, this paper will end with a brief conclusion in Section V. II. DOM P IXEL A. Fluorescence Lifetime Measurement Method When a very short pulse (typically less than 100 ps) excitation light irradiates a specimen which carries fluorescent probes, the probes will emit fluorescence with rapid exponential decaying, typically nanosecond to several tens of nanoseconds, as shown in Fig. 1(a). When the fluorescence intensity becomes 1/e of the initial intensity A0 at t = τ , the time τ is defined as the lifetime of the fluorescence [10]. The fluorescence intensity A(t) is proportional to the population of fluorophores in the excited state. The fluorescence decaying is detected in the system shown in the block diagram in Fig. 1(b). Fluorescence incidents on the photodetector and generates photoelectrons. Ideally, the electron generation in the photodetector follows the fluorescence decay. The number of generated electrons per unit time can be expressed as t − t0 n(t) = n0 · exp − τa

(1)

for t ≥ 0, where n0 is the initial electron number per unit time, t0 is the delay of photoelectron generation, and τa is the apparent or measured lifetime. In order to measure the fluorescence lifetime, a part of fluorescence signal electrons are detected and collected during the time windows. As described in Section II, the detection time window is set by a gate voltage VTD at the charge draining gate of photodiode. The detection time window in Fig. 1 is indicated by the duration when VTD is low level. Generated photoelectrons are accumulated by an electron collector every period T of the time window. The starting point of time window is delayed by a delay generator, and the electrons are collected from the time of td to td + T . If t0 is assumed to be zero for

Fig. 1.

Collection of fluorescence signal electrons.

simplicity, the number of accumulated electrons N (td ) by the charge collector is expressed as t d +T n0 τa exp − τtda , for (td ≥ 0) N (td ) = n(t)dt = n0 τa , for (td < 0) td (2) if T is chosen large enough so that the term of exp(−(td + T )/(τa )) is negligible. As the fluorescence from an actual biological specimen is very weak, the average number of collected photoelectrons for each excitation light is often less than one. To intensify the signal, the excitation is repeated many times, and the photoelectrons are repeatedly collected by the same window. The resulting total number of photoelectrons by the repeated process also follows (2). The dependence of the collected photoelectrons during the time window as a function of td is shown in Fig. 1(c), together with fluorescence photoelectrons and time window. The number of collected electrons first decays exponentially, followed by a long low level tail which is caused by noise fluctuation. Fluorescence lifetime is obtained from the gradient of the exponentially decaying region of the number of collected electrons, which is given by −1/τa , as shown in Fig. 1(c). B. Pixel Structure and Operation To implement the windowed detection of fluorescence photoelectrons, a pixel structure using DOM is proposed. To compare

LI et al.: CMOS IMAGE SENSOR WITH DOM PIXELS FOR FLUORESCENCE LIFETIME IMAGING

Fig. 2.

2717

Concept of TSCT pixel and DOM pixel. Fig. 3. Layout of the whole pixel.

the DOM pixel with the TSCT pixel, conceptual illustrations of their layout and potential profile are shown in Fig. 2. The DOM pixel, as shown in Fig. 2(b), removes the transfer gate between the PPD and PSD in the TSCT pixel. In the TSCT pixel shown in Fig. 2(a), a small potential barrier may be created at the edge of the transfer gate, which causes a loss of sensitivity and even a black defect because of short time window. The DOM pixel has a simplified structure, which leads to higher fill factor (higher sensitivity) of photodiode. The charge transfer delay caused by electrons trapped by Si−SiO2 under the TX01 gate can also be avoided. A charge draining gate (TD gate) is attached beside the carrier path (channel) from the PPD to the PSD in the DOM pixel. By closing the TD gate, the charges are transferred to the PSD. By opening the TD gate, the charges in the PPD are drained before they move to the PSD. A potential barrier may be created if the channel near the TD gate is not well engineered; however, it does not cause a black defect pixel. The operation for reading out the signal in the PSD is done by the same manner with the TSCT pixel using the TX gate. The DOM pixel in Fig. 2(b) is realized in the layout shown in Fig. 3. The actual shape of the PPD in the layout in Fig. 3 has a few steps to gradually widen the channel width from aperture area to the PSD, rather than a continuous slope in Fig. 2. This is to simplify the layout design, because the device fabrication technology does not allow using a diagonal line other than that with 45◦ or 135◦ . The DOM pixel uses a standard CMOS image sensor technology with PPD option. Pinned diode structures for both charge generation and storage have the advantages of low dark current and no image lag [11]. Under a small electric field in a depleted pinned diode, the carrier transfer within a dimension of a few micrometers can be a few nanoseconds or less than 1 ns, as described in Section IV-B. This response time of subnanosecond is sufficient for observing time-resolved phenomena with subnanosecond resolution. The actual pixel consists of a PPD, a PSD, a TD gate, a TX gate between the PSD and floating diffusion (FD), a reset transistor, and a source follower amplifier with a pixel selection switch. The main part of the pixel is symmetrical along the A−A line, which contains two charge transfer paths, and the TD gate is situated beside each transfer path. This symmetrical structure allows the photosensitive area to be increased for high sensitivity.

Fig. 4. Operation for photocharge modulation and sequential readout in the DOM pixel.

To explain the charge modulation mechanism of the DOM pixel, the cross-sectional view of the pixel and potential profiles along the direction of the PPD to PSD (X−X ) and PPD to drain (Y−Y ) are shown in Fig. 4(a) and (b), respectively. When the TD gate is closed, monotonic positive lateral electric field is created for high-speed charge transfer from the PPD to the PSD, as shown by the dotted curve in Fig. 4(c). When the TD gate is opened, a potential dip is created in the carrier transfer path, as shown by the solid curve in Fig. 4(d). The carriers which drop into this potential dip are drained through the TD gate. The potential step is created under the TD gate generated by an additional p-type doping [12], [13]. This potential step is useful for reducing dark signal generated by a scooping effect of electrons in the drain region, when repeating the TD closing and opening. The operation for reading out the fluorescence-generated electrons is shown in Fig. 4(e). After the repeated windowed signal detection, the signal stored in the PSD is read out using

2718


a true correlated double sampling (CDS) operation. To do this, the FD is first reset, and then, the charge in the PSD is transferred to the FD by the readout process, while the TD is always opened for draining the dark electrons in the PPD. C. Windowing Using DOM Pixel A time-windowed fluorescence signal can be detected in the following process. During the light pulse excitation, the TD gate is opened to drain unwanted charges directly generated by the excitation light. Then, fluorescence light with exponential decaying is emitted, and it generates signal charges. The TD gate remains open until the beginning of the time window. During the time window from td1 to td1 + T , the TD gate is closed. Therefore, a part of the decaying signal charges generated in the PPD are transferred to the PSD. This process is repeated many times using the delayed time windows with the same relative delay to the excitation light pulse, to intensify the weak fluorescence signal. To detect the signal electrons in a different time window, another delay time of td2 is used, as shown in Fig. 1(c). III. D ESIGN OF A T IME -R ESOLVED I MAGER U SING DOM P IXELS A. Imager Architecture Fig. 5(a) shows the block diagram of the CMOS FLIM chip, and Fig. 5(b) shows the timing diagram for the operation. The operation of the CMOS FLIM chip in one frame is divided into two parts: signal accumulation and readout. The timewindowed fluorescence signal detection is repeated during the accumulation time. The signal intensity can be controlled by the number of excitation light pulses or the number of capture windows. After the signal accumulation, all the pixel signals are read out using an X−Y addressing of pixels. The readout architecture and operation are the same as those of standard four-transistor active pixel CMOS imagers. For pixel noise reduction, an amplified CDS is performed at a column amplifier. The gain of the column amplifier can be set by the capacitance ratio, for a low gain of C1 /(C2 + C3 ) = 16/15 ∼ = 1.07 and a high gain of (C1 + C3 )/C2 = 30. A high gain of 30 is useful for low-noise signal readout if double-stage noise canceling is used [14]. To do this, two sample-and-hold capacitors CSH are used for the reset level and signal level of the high-gain amplifier output. The signals stored in CSH are horizontally read out by connecting them to two common horizontal signal lines and outputted using two output buffers. The final CDS is done by an off-chip differential-analog input ADC. B. DOM Pixel Design The pixel structure is optimized using a 3-D device simulator named SPECTRA. Fig. 6(a) shows the static simulation result of potential profile of the essential part of the pixel when the TD gate is closed and the TX gate is opened. The X and Y coordinates correspond to those shown in Fig. 3. The potential profile is suitable for barrierless charge transfer from the PPD to the PSD at high speed. Fig. 6(b) shows the simulated potential

Fig. 5. FLIM sensor chip block and operating timing. (a) Chip block diagram. (b) Timing diagram.

profile of the half pixel when the TD gate is opened and the TX gate is closed. The potential profile is suitable for draining all charges generated in the PPD at high speed, before charges reach the PSD. Simulation results show that the carrier transfer time within 2.8 μm from the PPD to the PSD is 0.18 ns, and the minimum electric field is 0.06 V/μm. During the charge draining from the PPD to the drain, the minimum electric field is 0.03 V/μm at the starting point of transfer, and the total transfer time within 1.8 μm is 0.05 ns. For time-windowed signal charge detection, unwanted charges should also be drained at high speed. If the potential barrier φB exists at the charge draining path in the channel region, as shown in the inset of potential profile along the Y−Y cross section in Fig. 7 (see also Fig. 4 for comparison), an unwanted photogenerated charge will be prevented from draining by potential barrier φB , once remains in the TD channel, and then will flow to the PSD when the TD gate is closed again. The unwanted charge is added to the timewindowed signal charge. The DOM pixel intends to collect photogenerated carriers during the TD gate is closed and to be unaffected by carriers when the TD gate is opened. However, because of imperfect light shielding in the PSD region, very slow-response carriers generated at the deep inside of silicon.


Fig. 6.

Three-dimensional potential profile of half pixel.

Fig. 8. Microphotograph of the chip prototype.

Fig. 7.

Simulated unwanted signal caused by barrier φB under TD.

Fig. 9. Experimental setup for fluorescence lifetime measurement.

Unwanted carriers generated by these reasons are superimposed on the pure time-windowed signal. The unwanted carrier signal changes the sensitivity of fluorescence signal. An unwanted signal caused by a given barrier φB is simulated, and the simulation result is shown in Fig. 7. The barrier φB is created by changing the TD gate voltage, as shown in the static simulation result at the right side of the y-axis in Fig. 7. The transient simulation result at the right side of the y-axis in Fig. 7 shows that the parasitic sensitivity can be suppressed to less than 10−2 for a TD gate voltage of larger than 3.5 V, where φB is smaller than 50 mV. The ratio of the highest unwanted signal intensity to the lowest unwanted signal intensity is defined as the extinction ratio, and the simulated extinction ratio is 394:1.

IV. I MPLEMENTATION AND E XPERIMENTAL R ESULTS

2719

the DOM pixels, an ADC, an FPGA, and a camera link interface chipset are implemented on an evaluation PCB. The control signal for excitation light is synchronized with the TD gate signal in the sensor. Excitation light irradiates on specimen, and fluorescence is emitted. Both the excitation light and fluorescence light are reflected toward the sensor by a mirror. The excitation light is filtered out by the optical bandpass filter. Only the fluorescence light can pass the spectral bandpass filter and is focused on the image sensor by a lens. The image sensor detects the fluorescence light. The detection time window as shown in Fig. 1 is set to 125 ns. The analog signal output of the sensor is fed to an ADC to be converted to a 14-b digital code. An FPGA is used for generating digital signals for controlling the sensor and receiving the digital code from the ADC. A camera link is used to transmit the digital code from the FPGA to a computer. After data processing, the decaying image is displayed on the monitor, and the data are stored in the computer.

A. Implemented Chip and Experimental Setup A image sensor chip with the DOM pixels is implemented using a 0.18-μm CMOS image sensor technology with PPD option. The microphotograph of the chip is shown in Fig. 8. The pixel array consists of 256 (row) × 256 (column) pixels, and the pixel size is 7.5 × 7.5 μm2 . A microlens array is formed on top of the pixel array. The experimental setup for fluorescence lifetime measurements is shown in Fig. 9. The time-resolved CMOS imager with

B. Basic Characteristics The unwanted signal intensity caused by the barrier φB is measured when the dc TD gate bias changes, and the measured result is shown in Fig. 10. The wavelength of the illuminated light is 440 nm. In this measurement, the number of electrons for a TD gate voltage of 0.6 V is 520 e− , and only 30 e− are drained at a TD gate voltage of 2.8 V. The number of unwanted electrons at a TD gate bias of 3.8 V is 15 e− in the PSD. The

2720


Fig. 12. Fig. 10. Measured unwanted signal intensity as a function of TD gate bias.

Fig. 11. (a) Averaged linearity at CDS gains of 1.07× and 30×. (b) Number of dark electrons increases with time window repetition.

measured unwanted signal intensity as a function of the TD gate bias in Fig. 10 agrees with the simulation in Fig. 7, except that a relative parasitic sensitivity of 0.03 is larger than the simulation result of less than 10−2 . One of the possible reasons is the stray light coming to the PSD through the structure above silicon. In each time window, the expected value of the number of photoelectrons by the weak fluorescence is often less than one. To evaluate the barrierless charge transfer for one electron from PPD to PSD, linearity at low illumination level is an important characteristic. Fig. 11(a) shows the number of accumulated electrons in the PSD as a function of the number of incident light pulses. The signal is averaged using an output of 10 × 10 pixels near the center of pixel array. The linearities for CDS gain settings of 1.07× and 30× are shown. The light pulse is generated by a 440-nm laser with a pulsewidth of 100 ps. The measured conversion gain is 30 μV/e− . When the CDS gain is 1.07×, the saturation level determines the full well capacity of the PSD, which is 3800 e− . A CDS gain of 30× is used for amplifying weak fluorescence signals below 1300 e− , and a maximum input referred signal of 1300 e− (39 mV at the source follower output) is determined by the saturation level of the CDS amplifier circuits. The linearity of the 10 × 10 pixels at a CDS gain of 30× is plotted in Fig. 12. Fig. 12(a) shows the linearity when the number of light pulses increases from 0 to 1000. Fig. 12(b) shows the magnified plot of Fig. 12(a) when the number of light pulses increases from 0 to 100, which shows that one electron is generated in the PPD and transferred to the PSD every 20 light pulses. All the pixels have good linearity for the signal level of even below one electron.

Linearity of 10 × 10 pixels at a CDS gain of 30×.

The DOM pixel with high-speed TD gate clocking may cause dark electrons scooped from the drain region via traps, which exist in Si−SiO2 interface under the TD gate. During each time window which has a period of 250 ns, a part of the trapped charges are transferred to the PSD when the TD gate is closed, and the accumulated dark charges increase linearly when the number of time window repetitions increases, as shown in the measurement results in Fig. 11(b). The duty of the TD pulse is set to 50% in the measurement. The generation ratio of dark electrons to the number of applied TD gate pulses is one electron per 546 pulses. This dark signal will cause an offset when detecting light signal and also shot noise. For all measurements in this paper, the number of time windows is set to be a constant value, and offsets due to the dark electrons are subtracted from the signal output. C. Fluorescence Lifetime Measurements Fig. 13 shows the measured signal intensity decaying curves for fluorescence emission of four types of fluorophores and intrinsic response of the time-resolved imager. An ultraviolet laser diode with a wavelength of 374 nm and a pulsewidth of 80 ps is used for all the measurements. The DOM pixel with a finite charge transfer time has its own intrinsic lifetime τ0 , which is determined by the dispersion of light spot on the photodiode and the dispersion of the transfer time. In the measurement of intrinsic lifetime, excitation light is illuminated directly on the sensor chip. In Fig. 13, the horizontal axis is the delay time of the time window, which is shifted from −20 to 80 ns with a time step of 1 ns. The name and center wavelength of the four kinds of fluorophores are shown in Fig. 13. Aside from the rapid decay region, two decay curves of intrinsic response and blue acrylic screen have another slow decay region. In the intrinsic response, direct laser illumination of the chip can excite parasitic fluorescence from packaged chip materials such as microlens. To prevent the fluorescence emission from the packaged chip materials, 374-nm laser light is filtered out using the optical bandpass filter. However, in the case of blue acrylic screen, the fluorescence itself may excite the chip materials because the fluorescence has a wavelength of 420 ∼ 460 nm using the optical bandpass filter. Fig. 14 shows the comparison of the target lifetime and measured lifetime in Fig. 13. The device intrinsic lifetime τ0 is measured to be 2 ns. The target lifetime is measured by an accurate fluorescence lifetime measurement tool using the TCSPC method. As the target lifetime increases, the influence


Fig. 13. Lifetime measurement for sensor intrinsic response and four types of fluorophores.

2721

Fig. 15. Decay image of two fluorophores of different lifetimes [(left side) orange acrylic screen, τa = 7.3 ns; (right side) green acrylic screen, τa = 4.4 ns]. TABLE I S UMMARY OF P ERFORMANCES OF THE P ROTOTYPE S ENSOR C HIP

Fig. 14. Comparison of measured lifetime and ideal lifetime.

of τ0 becomes smaller. If the target lifetime is comparable or smaller than τ0 , the measured lifetime is lower limited by τ0 , and the error of the measured lifetime from the target becomes relatively large. Measured fluorescence decaying is a convolution of the target fluorescence decaying and the time response of the DOM pixel. If the time response of the DOM pixel is known, the resulting target fluorescence decaying and the resulting target lifetime can be estimated by a deconvolution technique [15]. This is left as a future study. The resolution of the lifetime measurement is dependent on the signal amplitude, photon shot noise, and readout noise. It can be estimated with an analysis similar to that for the time-of-flight range finder [16]. The intensity decay images of two types of fluorophores are shown in Fig. 15. An orange acrylic screen, which is put on the left side in each subpicture, has a measured lifetime of 7.3 ns and decays obviously slower than the green acrylic screen, which is shown on the right side with a measured lifetime of 4.4 ns. The delay time is shown at the bottom of each image. After 40 ns, the intensity decreases to less than 1% of the initial intensity, so the fluorescence images in the two sides become dark. Some parasitic fluorescence decays much slower than the two fluorophores, causing a bright spot which appears at the left side of the image. Performances of the prototype sensor chip with DOM pixels are summarized in Table I.

V. C ONCLUSION A time-resolved CMOS image sensor for fluorescence lifetime imaging has been presented. A prototype sensor using DOM pixels has been implemented and tested. The barrierless charge transfer structure can transfer one electron from the PPD to the PSD at high speed, which is proven by the linearity measurement result. The fluorescence decaying has been successfully measured with 1-ns time step, using the prototype sensor. The prototype sensor is useful for a compact low-cost camera for FLIM in biological measurements. Further investigation is necessary for reducing the device lifetime for more accurate lifetime measurement of subnanosecond timescale and the dark signal generation effect due to the scooping of electrons in drain region. R EFERENCES [1] D. Elson, S. Webb, J. Siegel, K. Suhling, D. Davis, J. Lever, D. Phillips, A. Wallace, and P. French, “Biomedical applications of fluorescence lifetime imaging,” Opt. Photon. News, vol. 13, no. 11, pp. 26–32, Nov. 2002. [2] K. Yoshiki, H. Azuma, K. Yoshioka, M. Hashimoto, and T. Araki, “Finding of optimal calcium ion probes for fluorescence lifetime measurement,” Opt. Rev., vol. 12, no. 5, pp. 415–419, Sep./Oct. 2005. [3] X. F. Wang and B. Herman, Fluorescence Imaging Spectroscopy and Microscopy. New York: Wiley, 1996. [4] Y. Maruyama and E. Charbon, “A time-gated 128 × 128 CMOS SPAD array for on-chip fluorescence detection,” in Proc. Int. Image Sens. Workshop, Hokkaido, Japan, Jun. 2011, pp. 270–273. [5] S. Bellisai, F. Guerrieri, S. Tisa, and F. Zappa, “3D ranging with a singlephoton imaging array,” in Proc. SPIE—IS&T Electronic Imaging, San Fransisco, CA, Jan. 2011, vol. 7875, pp. 787 50M-1–787 50M-6.

2722


[6] F. Guerrieri, S. Tisa, A. Tosi, S. Bellisai, B. Markovic, and F. Zappa, “Linear arrays of single-photon detectors for photon counting and timing,” in Proc. SPIE—IS&T Electronic Imaging, San Fransisco, CA, Jan. 2011, vol. 7875, pp. 78750N-1–78750N-9. [7] H. Yoon, S. Itoh, and S. Kawahito, “A CMOS image sensor with in-pixel two-stage charge transfer for fluorescence lifetime imaging,” IEEE Trans. Electron Devices, vol. 56, no. 2, pp. 214–221, Feb. 2009. [8] H. Takeshita, T. Sawada, K. Itoh, T. Iwahori, and S. Kawahito, “Buried photodiode structure for high-speed charge transfer,” in Proc. ITE Annu. Conv., Kyushu, Japan, Aug. 2008, (in Japanese). [9] C. Tubert, L. Simony, F. Roy, A. Tournier, L. Pinzelli, and P. Magnan, “High speed dual port pinned-photodiode for time-of-flight imaging,” in Proc. Int. Image Sens. Workshop, Bergen, Norway, Jun. 2009. [10] J. R. Lakowic, Principles of Fluorescence Spectroscopy, 3rd ed. New York: Springer-Verlag, 2006. [11] N. Teranishi, A. Kohno, Y. Ishihara, and K. Arai, “No image lag photodiode structure in the interline CCD image sensor,” in Proc. IEDM, Dec. 1982, pp. 324–327. [12] J. Hynecek, “Virtual phase technology: A new approach to fabrication of large-area CCDs,” IEEE Trans. Electron Devices, vol. ED-28, no. 5, pp. 483–489, May 1981. [13] K. Yasutomi, S. Itoh, and S. Kawahito, “A two-stage charge transfer active pixel CMOS image sensor with low-noise global shuttering and a dualshuttering mode,” IEEE Trans. Electron Devices, vol. 58, no. 3, pp. 740– 747, Mar. 2011. [14] N. Kawai and S. Kawahito, “Noise analysis of high-gain, low-noise column readout circuits for CMOS image sensors,” IEEE Trans. Electron Devices, vol. 51, no. 2, pp. 185–194, Feb. 2004. [15] D. Connor, W. Ware, and J. Andre, “Deconvolution of fluorescence decay curves, A critical comparison of techniques,” J. Phys. Chem., vol. 83, no. 10, pp. 1333–1343, Mar. 1979. [16] R. Lange and P. Seitz, “Solid-state time-of-flight range camera,” IEEE J. Quantum Electron., vol. 37, no. 3, pp. 390–397, Mar. 2001.

Zhuo Li (S’10) received the M.E. degree from Shizuoka University, Hamamatsu, Japan, in 2009, where he is currently working toward the Ph.D. degree. His research interest is in time-resolved CMOS image sensors.

Shoji Kawahito (S’86–M’88–SM’00–F’09) received the Ph.D. degree from Tohoku University, Sendai, Japan, in 1988. Since 1999, he has been a Professor with Shizuoka University, Hamamatsu, Japan.

Keita Yasutomi (S’08–M’11) received the Ph.D. degree from Shizuoka University, Hamamatsu, Japan, in 2011. He is currently an Assistant Professor with Shizuoka University. His research interests include CMOS image sensors and pixel design.

Keiichiro Kagawa (M’11) received the Ph.D. degree from Osaka University, Osaka, Japan, in 2001. Since 2011, he has been an Associate Professor with Shizuoka University, Hamamatsu, Japan.

Juichiro Ukon received the M.S. degree from Kansai University, Osaka, Japan, in 1978. Since 1980, he has been with Horiba, Ltd., Kyoto, Japan, where he has developed laser application measurement systems and optical spectroscopic instruments.

Mamoru Hashimoto received the Ph.D. degree in basic science from the University of Tokyo, Tokyo, Japan, in 1997. Since 1997, he has been with Osaka University, Osaka, Japan, where he is currently an Associate Professor.

Hirohiko Niioka received the Ph.D. degree in frontier bioscience from Osaka University, Osaka, Japan, in 2009. Since 2009, he has been an Assistant Professor with Osaka University. His research interest is constructing electron microscopes.