Performance of chip-size wavelength detectors

Performance of chip-size wavelength detectors Oliver Schmidt, Peter Kiesel *, Michael Bassler Palo Alto Research Center Incorporated, 3333 Coyote Hill Rd., Palo Alto, CA 94304 * Corresponding author: peter.kiesel@parc.com Abstract: Chip-size wavelength detectors are composed from a linear variable band-pass filter and a photodetector array. The filter converts the incident spectral distribution into a spatial distribution that is recorded by the detector array. This concept enables very compact and rugged spectrometers due to the monolithic integration of all functional components on a single chip. This type of spectrometer reveals its most convincing advantages through appropriate systems integration. We discuss the advantages of this concept for spectroscopy of light distributions that are hard to focus onto the entrance slit of a conventional spectrometer, namely large light emitting areas and moving point-like light sources. The excellent spectral performance of the system is demonstrated for both light input geometries. 2007 Optical Society of America OCIS codes: (300.6190) Spectrometers, (310.6860) Thin films, optical properties. References and links 1. N. Damean, S. K. Sia, V. Linder, M. Narovlyansky, and G. M. Whitesides, Space and time-resolved spectrophotometry in Microsystems, PNAS 102, 10035-10039 (2005). 2. S. Grabarnik, R. Wolffenbuttel, A. Emadi, M. Loktev, E. Sokolova, and G. Vdovin, Planar double-grating microspectrometer, Opt. Express 15, 3581 (2007). 3. P. Kiesel, O. Schmidt, S. Mohta, S. Malzer, and N. M. Johnson, Compact, low-cost, and high resolution interrogation unit for optical sensors, Appl. Phys. Lett. 89, 201113-1 - 201113-3 (2006). 4. N. M. Johnson, O. Schmidt, and S. M. Chokshi, Chip-Size Spectrometers (Palo Alto Research Center Incorporated, 2006). http://www.parc.com/research/projects/opticaldetectors/spectrometers.html. 5. F. W. Kavanagh, Spectral wedge interference filter combined with purifying filters, United States Patent 2,708,389 (1955). 6. N. Gat, Spectrometer apparatus, United States Patent 5,166,755 (1992). 7. J. A. Wahl, J. S. Van Delden, and S. Tiwari, Multiple-fluorophore-specie detection with a tapered fabryperot fluorescence spectrometer, Appl. Opt. 44, 5190-5197 (2005). 8. J. T. Olesberg, C. Cao, J. R. Yager, J. P. Prineas, C. Coretsopoulos, M. A. Arnold, L. J. Olafsen, and M. Santilli, Optical microsensor for continuous glucose measurements in interstitial fluid, SPIE Proc. 6094, 609403-1 - 609403-10 (2006). 9. O. Schmidt, M. Bassler, P. Kiesel, C. Knollenberg, and N. Johnson, Fluorescence Spectrometer-on-afluidic-chip, Lab Chip 7, 626-629 (2007). 10. O. Schmidt, P. Kiesel, S. Mohta, and N.M. Johnson, Resolving pm wavelength shifts in optical sensing, Appl. Phys. B 86, 593 (2006). 1. Introduction In recent years there has been a tremendous interest in developing sensor systems for health care service, industrial process monitoring and environmental monitoring. Typical measurements include proving the presence or absence of an analyte, determining the quality of an industrial process or monitoring the reaction or binding dynamics of an analyte in water, blood, aerosols, air, food, and other specimens. Optical spectroscopy can be extremely sensitive, selective and can be used for continuous real-time monitoring without contaminating the sample. Thus, they are used frequently in sensor applications. In general it is desirable to design optical systems that are much more (C) 2007 OSA 23 July 2007 / Vol. 15, No. 15 / OPTICS EXPRESS 9701

compact and rugged than what is currently available. Semiconductor laser diodes and light emitting diodes (LED) have helped to integrate powerful and reliable light sources into optical systems. Optical fibers and waveguide structures have drastically reduced alignment issues of optical systems and allow for compact and flexible systems. Still lacking for truly powerful optical systems are compact spectrometers that can be easily integrated in order to leverage the huge potential of the spectral dimension of such systems. Damean et al. [1] included a diffraction grating in their microfluidic chip in order to achieve on-chip wavelength separation but still relied on a bulky bright-field microscope to detect the spectrum. Final goal has to be a complete integration of the spectrometer function. Prism and grating based approaches are somehow problematic since they intrinsically need a distance between the dispersion and detection element for wavelength discrimination. This limitation has been addressed recently by clever design, thereby minimizing the volume of the spectrometer [2]. 2. Concept of chip-size wavelength detectors In this paper we discuss a concept for a wavelength detector that replaces the dispersive element by a spatially selective filter. The concept is based on a linear variable band-pass filter (LVF) which selectively transmits light at different locations depending on the lights wavelength and photo-detector arrays which record the light intensity behind the filter [3,4]. The filter is attached directly onto the photo-detector chip and can even be deposited directly onto the chip. The resulting detector provides compactness, robustness, and ease of fabrication. It leverages existing system components and manufacturing techniques and does not exceed the size of the photo-detector array chip. It has no moving parts and enables monolithic integration into optical systems. At the same time, it features high performance and flexibility because the filter is customizable for high spectral resolution or broad spectral range. A wide range of photo-detectors can be used for high sensitivity or fast read-out speed. The entire spectral range from the ultraviolet and visible range to the infrared spectral region can be addressed. Existing photo-detector arrays based on silicon, germanium, InAs, InGaAs or PbTe, or even bolometer arrays can be used. The basic, underlying concept was introduced many decades ago but was not implemented in many applications due to the lack of cheap and powerful detector arrays [5,6]. In recent years the topic has received new interest due to the demand for cheap and compact spectrometers [7,8]. Even so, chip-size spectrometers are still not significantly used because existing systems have not addressed the specific requirements and challenges of this technique that arise primarily from the unique light input geometry. Most applications try to simply replace a conventional spectrometer by a chip-size wavelength detector and do not take advantage of the distributed detection area of these devices, which is especially favorable for spectral characterization of large light emitting objects or moving particles as will be demonstrated in this paper. Further, many designs neglect the need for uniform large-area illumination of the detector. Another key issue, which has to be taken into account, is the angle dependence of the LVF to the incident light. In the following, these specific issues of the device are addressed. Design strategies and calibration procedures are described. Finally, it is demonstrated that a chip-size spectrometer can resolve a spectrum with the same accuracy as a conventional spectrometer. The photosensitive area of the wavelength detector is determined by the size of the LVF and the photo-diode array. The entire detector needs to be illuminated in order to benefit from the entire spectral range of the spectrometer. This operating mode is fundamentally different from conventional spectrometers where the light is focused onto a small slit in order to receive good wavelength separation (Fig. 1(a)). This makes a comparison of both spectrometers very difficult because it truly depends on the dimension and divergence of the incident light beam. In general, it appears that a grating based spectrometer is favorable for fixed point light sources which can be easily imaged onto the slit whereas the chip-size spectrometer is especially favorable for large light emitting areas such that no or only simple optics are necessary to direct the light onto the detector (Fig. 1(b)). In fact, the wavelength detector could be attached directly to the light emitting area. This allows a very simple installation (C) 2007 OSA 23 July 2007 / Vol. 15, No. 15 / OPTICS EXPRESS 9702

without critical alignment in addition to the advantage that the overall system is very compact and robust. Of course there are drawbacks of this technique. The detector determines the spectral information of the light source laterally, that is, by measuring different spectral components at different locations on the detector. Adverse effects on the spectrum have to be avoided either by illuminating the entire detector surface with a constant light intensity or by correcting for an inhomogeneous illumination. grating-based monochromator LVF CCD sensor LVF CCD sensor slit (a) Light source (b) (c) Fig. 1: Light input geometries of a grating based spectrometer (a) in comparison to a chip-size spectrometer. The chip-size spectrometer is particularly valuable for (b) large area light sources and (c) moving point-sources A special case is a moving light emitting object. Recording such a moving object with a long integration time effectively renders the point source into a large area light source (Fig. 1(c)). The geometry of the detector enables an optical detection method that can even take advantage of the motion of an optical source. The spectral information of the lightemitting object is gathered step-by-step while it is moving across the wavelength detector. A lab-on-a-chip system was proposed, which is designed to record the fluorescence spectrum of particles on-the-flow [9]. The particles are continuously excited by the excitation laser and emission spectra are recorded while the particle is traversing the detector. 3. Description of prototype A compact spectrometer was realized by mounting a broad-band LVF that covers the entire visible spectral range in front of a 16bit CCD camera. The distance between the filter and the CCD sensor was approximately 10mm due to the design of the camera. The camera had a resolution of 1024 256 pixels with a pixel size of 26 26 µm 2. The exposure time could be varied between 30 µs and 4 s. The LVF covered the spectral range from 400 to 700 nm with a FWHM of the transmission peak of ~1% of the transmission peak wavelength and a spectral gradient of about 30 nm/mm. The resolution of the spectrometer is mostly determined by the FWHM of the LVF as described in detail elsewhere [10] which can be realistically designed between 1 and 10 nm in the visible spectral range. Note, that a smaller FWHM will increase the spectral resolution for trading light efficiency. This is similar to grating based spectrometers where changing the entrance slit width has the same effect. For spectral calibration purpose, light from a Halogen lamp was spectrally filtered by a monochromator and coupled into a polymer optical fiber. At the opposite end of the fiber the light was collimated and directed onto the detector such that the whole detector was illuminated similar to the situation depicted in Fig. 1(b). Due to the special transmission properties of the LVF monochromatic light propagates through the filter at a single sharp position, whereas the filter blocks the light at all other positions. By changing the wavelength of the incident light we correlate each position of the detector with a specific wavelength band, the width of the band given by the FWHM of the LVF. In order to achieve valid intensity values for all wavelength bands we need to account for the spectral dependency of the detector sensitivity C cam ( and the transmission function of the (C) 2007 OSA 23 July 2007 / Vol. 15, No. 15 / OPTICS EXPRESS 9703

LVF C LVF ( (both shown in Fig. 2(a)). For the LVF used in this experiment the transmission peak height and the FWHM increase almost proportionally with wavelength. Another important aspect for achieving an accurate spectrum is the correction for inhomogeneous illumination of the spectrometer. In practice it is quite challenging to assure homogeneous illumination. Therefore, we introduce referencing regions in the spectrometer which monitor the intensity distribution across the detector independently from the spectrum of light. In the present work we have realized this intensity referencing technique by doing two sequential measurements of the same light distribution. The first measurement is done with the filter mounted on top of the camera, such that the intensity distribution behind the filter yields spectral information about the incident light I raw (. The second measurement is done without the filter in order to determine the intensity distribution of the incident light I in (. Note, that the spatial intensity information I in (x) has been converted into a function of wavelength I in ( under consideration of the calibration procedure described above. Consequently, the correct spectral distribution I( can be calculated as follows: Iraw( I( = Ccam( CLVF ( Iin( (1) In a final implementation, one or more reference lines have to be integrated into the filter in order to allow measuring I raw ( and I in (x) simultaneously. This can be achieved by leaving a line within the filter uncoated such that light propagation through this uncoated line does not depend on the wavelength of light. For optimum performance, reference lines should be positioned in close proximity to the LVF line. calibration factor 1.0 0.8 0.6 0.4 0.2 LVF correction Camera response Reference 400 450 500 550 600 650 700 (a) wavelength (nm) intensity 1.2 raw data calibrated data reference measurement 1.0 0.8 0.6 0.4 0.2 0.0 400 450 500 550 600 650 700 (b) wavelength (nm) Fig. 2: (a) Calibration factors for wavelength dependency of LVF and camera as well as correction for inhomogeneous illumination. Inset: Camera snapshot without LVF (reference). (b) Snapshot and intensity profile with LVF. The calibrated intensity profile of light distribution behind the camera yields spectrum of incident light (white LED). The spectrum is in perfect agreement with measured data from a grating based spectrometer. 3.1 LED spectroscopy In order to demonstrate the performance of the wavelength detectors experimentally we analyzed the spectrum of the white LED which is fabricated from a blue LED that was coated with a yellow phosphor layer. The LED light was first focused onto a grating based monochromator in order to measure the reference spectrum as depicted in Fig. 1(a). The monochromator was calibrated beforehand with a calibrated light source such that the wavelength dependent grating efficiency and the detector sensitivity were eliminated from the measurement. Figure 2(b) shows the spectrum of the LED after calibration. The spectrum (C) 2007 OSA 23 July 2007 / Vol. 15, No. 15 / OPTICS EXPRESS 9704

features a sharp peak around 460 nm from the blue pump LED and a much broader distribution centered near 570 nm which originates from the phosphorescence. Next, the LED light was collimated by using an achromatic lens to create a beam with a diameter of approximately 15 mm. The light was directed onto the chip-size wavelength detector described above. Figure 2(b) shows a snapshot taken by the camera and the intensity profile that was extracted from this image by averaging over 20 lines (this corresponds to approximately 50% of the filter width). The intensity profile was corrected by a reference profile as described above that was recorded beforehand without the LVF in place. Figure 2(a) includes the corresponding snapshot and intensity profile that was extracted as described above. The shape of the intensity profile reflects the shape of a Gaussian light spot that was created by the lens of our experimental setup. In Fig. 2(b) we compare the intensity profile before and after correcting the data with the calibration factors as well as with the spectrum received from the grating-based spectrometer. The obtained spectrum is in perfect agreement with the grating based measurement. 3.2 Spectroscopy of a moving LED In another experiment we focused the light of the white LED and moved the resulting spot across the LVF by moving the LED/lens combination with respect to the chip-size spectrometer (Fig. 1(c)). Thereby, we tested the performance of the chip-size spectrometer under a completely different light input condition than considered previously. In this scenario only a small spectral region is sampled at a time and the entire spectrum is recorded sequentially. The spectral region that is sampled is defined by the position and the size of the light spot on the LVF. Figure 3(a) shows the intensity profile that is recorded by the CCD sensor for four different spot positions. The profiles are relatively broad because of the diverging light beam between the LVF and the sensor as depicted in Fig. 1(c). Still, the intensity of the profiles can be attributed to a well-defined wavelength because the light beam sampled only a small region of the LVF. Therefore, we integrated the intensity of the profile and ascribed the entire intensity to the wavelength that is determined by the position of the light spot on the LVF. We determined this position by the peak location λ peak of a Gaussian curve that is fitted to the profile (see Fig. 3(a)). The intensity was integrated from λ peak w to λ peak +w where w is the FWHM of the Gaussian fit. 30 25 30 25 moving LED spectrum static LED spectrum 20 20 intensity 15 intensity 15 10 10 5 5 (a) 0 400 450 500 550 600 650 700 750 wavelength (nm) (b) 0 400 450 500 550 600 650 700 wavelength (nm) Fig. 3: (a) Sequentially recorded intensity profiles of a focused and moving LED (see Fig. 1(c)). The peak position of each curve is correlated with a spectral region that is defined by the LVF s transmission properties. The area below the curve quantifies the intensity of the corresponding spectral component. (b) Combining the spectral band and intensity information of each profile results in the spectrum of the LED. The spectrum of the moving LED is compared to the spectrum of the static experiment (before calibration). (C) 2007 OSA 23 July 2007 / Vol. 15, No. 15 / OPTICS EXPRESS 9705

The light spot was shifted in increments of 0.13 mm along the gradient of the LVF and the resulting intensity profiles were recorded. For a filter length of 10 mm this resulted in about 80 curves. For each profile we determined the integral and assigned it to the corresponding wavelength as described above. The resulting data are shown in Fig. 3(b) and compared to the spectrum that was obtained from the static large-area spectroscopy approach described above. Excellent agreement was achieved between the two light input schemes. The small deviations are due to systematic experimental errors. As mentioned above, the design of the CCD camera required a gap between LVF and CCD chip. In the above experiment we chose to position the focal point of the optical setup on the filter. The results would have changed dramatically if we had put the focal point onto the detector. In this case the incident light samples many positions on the LVF and it is impossible to separate the different wavelength components because the entire light intensity is detected by a single spot on the camera. Thus, the features of the spectrum would be washed out essentially because the transmission properties of the filter are averaged over a large area. On the other hand, if we place the focal point somewhere in front of the filter (see spotted line in Fig. 1(c)), we achieve a situation where we can take advantage of the divergent light beam in order to measure multiple spectral regions at once. In this case, many positions of the filter receive light at a well-defined angle that is determined by the lateral position of the focal point and the distance from the filter. If we assigned the entire intensity that is received at the CCD sensor to a single wavelength, we would again be averaging the transmission properties of many positions and angles and end up with a washed out spectrum. Fortunately, we can separate the different filter locations in this geometry, because the light that transmits at a particular location through the filter does not mix with the light from a different location. Each pixel of the CCD receives light only from a small filter region and within a small angular spectrum. 4. Conclusion Chip-size wavelength detectors were designed by combining linear variable band-pass filters with a CCD camera. The filter converts the spectral information of the incident light into a spatially dependent signal that is analyzed by the detector. The filters can be designed either to cover a broad spectral range or to enable high wavelength resolution. We have demonstrated a spectrometer for the visible spectral range and have discussed calibration and referencing techniques. Inclusion of a reference line that monitors the intensity distribution of the incoming light is essential for this kind of wavelength detector to eliminate errors from inhomogeneous illumination. In addition, it is essential to correct the measurement with regard to the angle of the incident light. Compact broad-band spectrometers enable integration of spectroscopic techniques onto lab-on-a-chip devices. Due to their extended detection area, chip-size spectrometers are especially favorable for large light-emitting areas or moving particles. We recorded the spectrum of a LED under both light-input geometries, and in both scenarios we achieved excellent agreement with a reference measurement obtained from a conventional spectrometer. Acknowledgments The authors are pleased to acknowledge helpful discussions with Noble M. Johnson (PARC) and Gottfried Döhler (Max Planck Research Group, Institute of Optics, Information and Photonics, Erlangen, Germany). The research was partly funded by ONR under grant N00014-05-C-0430 monitored by Jeremy Walker, Susan Rose-Pehrsson, and Paul Armistead. (C) 2007 OSA 23 July 2007 / Vol. 15, No. 15 / OPTICS EXPRESS 9706