Low-noise CMOS Fluorescence SensorDavid Sander, Marc Dandin, Honghao Ji, Nicole Nelson, Pamela AbshireDepartment of Electrical and Computer Engineering Institute for Systems ResearchUniversity of Maryland, College Park, Maryland 20742 USAEmail: {dsander, mpdandin, jhonghao, nmnelson, pabshire}@umd.eduAbstract— This paper reports a novel integrated circuit forfluorescence sensing. The circuit implements a differential read-out architecture in order to reduce the overall noise figure. Thecircuit has b een fabricated in a commercially available 0.5 µmCMOS technology. Preliminary results show that the reset noiseis reduced by a factor of 1.42 and the readout noise by a factor of9.20 when the pixel is operated in differential mode versus single-ended mode. Spectral responsivity characteristics show that thephotodiodes are most sensitive at 480 nm. Using a commerciallyavailable emission filter, the sensor was able to reliably detect aconcentration of Fura-2 as low as 39 nM. The sensor was usedto perform ratiometric measurements and was able to reliablydetect a free calcium concentration of 17 nM.I. INTRODUCTIONFluorescence sensing is a mature technology commonlyused in cell biology. For many analytes it provides thehighest sensitivity and selectivity available. The ability todetect fluorescence on-chip i s an important feature of lab-on-a-chip devices. Several groups [1]–[5] are pursuing integratedfluorescence sensing for applications ranging from DNAanalysis to pathogen detection. The devices demonstrated inthese works make use of conventional detector architectures(reverse biased p-n and p-i-n junctions) for transduction offluorescence. However, achieving high-sensitivity fluorescencedetection requires high quality optical filters as well as highquality optical detectors. In particular, detector noise must bereduced below the signal level to be detected (usually less than10 photons/µm2·s).In this paper, we propose an active pixel sensor architecturein which the photocurrent is measured differentially. Experi-mental results demonstrate that reset and readout noise areboth reduced when the pixel is operated in this manner. Thedetector is a p+/nwell junction which offers high sensitivity inthe blue to green region of the electromagnetic spectrum. Thismakes the sensor ideal for use in a wide range of fluorescenceassays, one example of which is monitoring calcium levelsusing the fluorophore Fura-2. The following sections describethe operation of the sensor and present results of a preliminaryanalysis on noise reduction in differential mode operation. Wethen present measurements and characterization of the spectralresponsivity of the detector along with demonstration of flu-orescence collection using an external macroscopic emissionfilter.II. SENSOR OPERATION AND NOISE ANALYSISThe detector is a 6-transistor differential active pixel sensorwith in-pixel sampling of the reset voltage (Figure 1). The resetVddi_gateresetCrstselectVrst VsigVdd VddVpbiasIn PixellightFig. 1. Differential pixel schematicvoltage is held on a capacitor and read out alongside the signalat the end of the integration period. By sampling the resetvoltage and reading out the signal differentially it is possiblein principle to virtually eliminate reset noise. Power supplynoise and other coupled noise sources will also be suppressedsubstantially. In practice the suppression of reset noise andreadout noise will be limited by charge injection and coupling.The optically active area is a 33.6 µm x 33.6 µm p+/nwellreverse biased diode where the nwell is tied to the powersupply Vdd. The p+/nwell junction helps to reduce noise bydecoupling the sensor from the substrate as well as s uppressingblooming effects. The hold capacitor is not required to belinear so it can be implemented with a MOSCAP or othernonlinear capacitance. In this circuit it has been implementedas a linear poly-poly capacitor for convenience, with nominalvalue of 20 fF.To examine the benefits of a differential readout with in-pixel sample and hold, we compare measurements from thesensor in single-ended mode to those in differential mode. Insingle-ended mode, the igate transistor is off for the entireexperiment, and the output of the sensor is measured withrespect to ground. Reset and select signals initialize the pixelbefore integration and select a row of pixels for readout afterresetselectIntegration Periodi_gateFig. 2. Timing diagram for differential sensorintegration as in a standard APS.For differential mode, three control signals are requiredto operate the pixel: select, reset, and igate (isolation gate).The isolation gate is on during the reset cycle and turns offimmediately before the end of the reset cycle. This minimizesnoise due to charge injection by providing a low impedancenode for dissipation of channel charges when both the isolationgate and reset gate close. Figure 2 shows the timing diagram.A series of 50 integration paths were measured for thesensor in single ended and differential mode of operation over3 orders magnitude of incident illumination. The illuminationsource was controlled using a monochromator and integratingsphere. The wavelength was 630 nm with 20 nm spread andoutput power fixed at 95.2 nW/mm2. A set of neutral densityfilters were used to decrease the optical power at the output ofthe integrating sphere. The neutral density filters were variedfrom optical density (OD) 2 to OD 5 in 0.5 increments. Wechose 630 nm because the neutral density filters are wellcharacterized for 630 nm light.The total variance of the measured reset voltage is the sumof reset noise and readout noise. Using our estimated readoutnoise, and the measured reset noise, we can estimate the truereset noise.V ar[V0] = σ2readout+ σ2reset(1)Readout noise was estimated as described below, using themethod outlined by Fowler [6]. The output voltage betweentwo successive measurements is equal to:V (i) = gQi+ Vnoise(i) − Vnoise(S1) (2)where g is the front end gain of the sensor, Qiis theaccumulated charge due to photocurrent and dark current, andVnoiseis the reset and readout noise at samples (S1) andi respectively. By subtracting successive measurements resetnoise is eliminated and we are left with an output which is afunction only of the photo-process and of the readout noise.Considered as stochastic processes, the photocurrent anddark current are Poisson processes, while the readout and resetnoise are assumed to be zero mean Gaussian processes.