566 IEEE SENSORS JOURNAL, VOL. 2, NO. 6, DECEMBER 2002 Imaging: Principles and Integrated Polarimeters Andreas G. Andreou and Zaven Kevork Kalayjian

Abstract—Polarization is a general descriptor of and con- of Huygens, Young, Fresnel, and Maxwell to lead us to the tains information about reflecting objects that traditional inten- understanding of the polarization of light that we have today; sity-based sensors ignore. Difficult computer vision tasks such as and, its study has lead to a deeper understanding of light and image segmentation and object orientation are made tractable with polarization vision techniques. Specularities, occluding contours, light-matter interaction. and material properties can be readily extracted if the Stokes po- Polarization is used to classify chemical isomers [22], to an- larization parameters are available. Astrophysicists employ polar- alyze solar and atmospheric phenomena [23], [24], to investi- ization information to measure the spatial distribution of magnetic gate stress and strain through photoelasticity, to classify mate- fields on the surface of the sun. In the medical field, analysis of the rials and recognize objects [25]–[30], and to analyze reflections polarization allows the diagnose of disease in the eyes. The retinae of most insect and certain vertebrate species are sensitive to polar- [31], [32]. Polarimetric devices and techniques have been in- ization in their environment, but humans are blind to this prop- vented to filter, separate, quantify, and modulate polarized light. erty of light. Biologists use polarimeters to investigate behaviors of Polaroid filters are used in magneto-optic drives, animals—vis-à-vis polarization—in their natural habitats. In this displays, and 3-D movie glasses. Polarization is used to track paper, we first present the basics of polarization sensing and then eye movements [33], and polarization microscopes help biolo- discuss integrated polarization imaging sensors developed in our laboratory. gists see things that would otherwise escape human perception [34]. Polarization measuring instruments today are either quasi- I. INTRODUCTION static and/or do not involve an imaging device (i.e. an array of ISUALIZING the polarization of light is beyond the ca- photosensitive sites with sensitivity to the state of polarization). V pabilities of the human eye, but it is an everyday behavior The rotating polarization analyzer type polarization imaging de- for many other creatures of the Earth. Bees use the polariza- vice, for example a liquid crystal polarization camera [35], [36] tion of scattered sunlight as a compass [1], [2]. Waterstriders, can give dynamic information in the millisecond time scale. backswimmers, and dragonflies use the polarization of reflected However, in the time between measurements the brightness can light to detect water and also under-water predators [3]–[7]. change, even if slightly, making it difficulty if not impossible to Both invertebrates (e.g. mantis shrimp [8], octopi [9], crayfish disentangle brightness variations from polarization information. [10], beetles [11], flies [12]) and vertebrates (e.g. rainbow trout For real-time, portable polarization imaging, an additional [13], [14], salmon [15], sunfish [16], [17], salamanders [18], and factor is the complexity of the data processing equipment and lizards [19]) are able to sense and use polarization in their envi- their size. We have fabricated and tested a series of experi- ronments. One species of polarization sensitive cephalopod, the mental prototypes for integrated CMOS polarization imagers cuttlefish, can modulate the reflectivity of its skin, thereby ma- [37]–[40]. In this paper we present a tutorial introduction to po- nipulating the reflected pattern of polarization for camouflage or larization phenomena and discuss a number of integrated sen- during copulation—like a polarization communication channel sory devices that have been engineered for polarization imaging [20]. As a biological visual modality, polarization is prevalent [41]. among animals. Although unable to naturally sense polarization, we humans II. POLARIZATION OF LIGHT:THE BASICS have still been able to measure and analyze polarization in our environment. In the history of science, the first recorded obser- Let us consider an electromagnetic wave propagating along vation of a polarization related phenomenon was made by the the z-axis, with the e-vectors confined to oscillate along only Danish professor Erasmus Bartolinus in 1669: the double-re- one perpendicular axis, either or . fraction of light through an Iceland spar [21]. It took the likes

(1) Manuscript received September 11, 2002; revised December 27, 2002. This work was supported by DARPA/ONR MURI N00014-95-1-0409 with Boston where and are the magnitude of the and components University on Automated Sensing and Vision Systems and by National Science Foundation grant EIA-0130812. The associate editor coordinating the review of of the electric field, and is the phase difference between these this paper and approving it for publication was Prof. R. W. Newcomb. two components. The appearance of in the above equa- A. G. Andreou is with the Department of Electrical and Computer Engi- tion allows for the most general case where orthogonal compo- neering, Johns Hopkins University, Baltimore MD 21218 USA. Z. K. Kalayjian is with Epigy Labs, Yerevan, Armenia. nents have different phase that may depend on spatial or tem- Digital Object Identifier 10.1109/JSEN.2003.807946 poral coordinates. Substitute for in each of the

1530-437X/02$17.00 © 2002 IEEE ANDREOU AND KALAYJIAN: POLARIZATION IMAGING: PRINCIPLES AND INTEGRATED POLARIMETERS 567

of the orthogonal components and their phase difference, the Stokes vector becomes

(5)

In order to fully describe the state of polarization of light, three (a) (b) (c) quantities must be known: , , and the phase difference Fig. 1. Polarization depends on the mode of oscillation of the electric field. . (a) Circularly polarized, (b) linearly polarized, and (c) unpolarized.

III. HOW DOES LIGHT BECOME POLARIZED? components in the equation and temporarily drop the explicit The origins of polarized radiation can be separated into two expression of the spatial and temporal dependence of in (1). categories: 1) radiation that is polarized upon emission from a polarized (2) light source; (3) 2) radiation that is originally unpolarized, but upon which the quality of polarization is subsequently imparted. By tracing the location of the tip of the electric field vector as It is interesting to note that the vast majority of polarized elec- the wave moves through space, we see that the resulting loci tromagnetic sources from category #1 are human made: lasers, ˇ describe a helix. synchrotron radiation, Cerenkov radiation, radio and television The ellipticity and orientation of the ellipse depend on the pa- broadcasts, and the Zeeman effect, to name some. Collision po- rameters . When the electric field oscillates in this larization in solar flares is one of the rare cases of a naturally mode, it is said to be elliptically polarized. In fact, if occurring polarized light source [24]. These phenomena share a , light is always elliptically polarized. The special case common trait—asymmetry or anisotropy (from a dipole, mag- where , and makes the ellipse into a netic field vector, or velocity vector) that results in a preferred circle, and the radiation is said to be circularly polarized. direction of oscillation of emitted radiation [43]. Another special case happens when and the Polarized radiation from category #2 above also depends on ellipse degenerates into a line. This is called linear polarization. an asymmetry. Unpolarized radiation, as was mentioned, ex- Depending on how the energy of the electromagnetic radiation is hibits no long-term preference for a particular oscillatory mode. distributed between the components , the electric field Consequently, in order for it to become polarized, some prefer- can be linearly polarized parallel to or . There is yet another ence has to be applied, via asymmetric work, to the unpolarized possible generally categorizable oscillatory mode that occurs radiation. This can be achieved through scattering, reflection, when and are not deterministically related. In other words, , or ; that latter two will be described in there is no consistent temporal relationship between the phases more detail below as they relate to the integrated polarization of the orthogonal electric fields. In this case, the e-vector takes a sensors discussed in this paper. random path as the wave travels through space. This radiation is A. Birefringence called natural or unpolarized. Fig. 1 shows the “head-on” view for circularly, linearly, and unpolarized light. In an isotropic medium, the electric field and the displace- All the possible states of polarization can be represented in ment vector are related by the scalar , the permittivity of the one vector known as the Stokes vector [42], named after George material. However, in an anisotropic medium, permittivity is a Gabriel Stokes, who introduced it in 1852. The Stokes vector tensor quantity is a shorthand notation that describes the polarization of light through four components (6)

The result of this tensor relationship in anisotropic crystals is (4) that and are not always parallel, which means that the di- rection of energy flow, or the ray direction, is not parallel to the wavefront normal, or the wave-normal direction. As a re- where is the total intensity (including any unpolarized por- sult, as the wavefront moves through the crystal, it is continu- tion), represents the magnitude of the e-vector component ally displaced in the ray direction as depicted in Fig. (2a). This along the or axes, represents magnitude of the e-vector displacement depends on the orientation of the e-vector with re- component at a 45 angle to the -axis, and represents right- spect to the axes of symmetry in the crystal. circularly polarized light. The stokes vector is also represented Calcite and Rutile are uni-axial crystals, which means there with alternative sets of symbols with the same respective mean- is one axis of the crystal, the optical axis, along which and ings: ,or . Rewritten in terms are parallel and no double-refraction occurs. If the e-vector 568 IEEE SENSORS JOURNAL, VOL. 2, NO. 6, DECEMBER 2002

Fig. 2. (a) When the electric field of incident radiation is parallel to a principle section of a birefringent crystal, like calcite, the wave-normal vector and the ray direction are not parallel. Consequently, the ray is displaced as it propagates through the crystal. (b) Unpolarized light entering a birefringent crystal will separate into two completely linearly orthogonally polarized rays. When these two rays exit the crystal, they are spatially displaced from one another, and the original ray appears as a double image. Figure adapted from [44]. oscillates in one of the principle planes of the crystal, which is a plane that contains the optical axis, then it sees an index of re- fraction , and the ray direction is oblique to the wave-normal direction. A beam of light that sees the index of refraction is called the extraordinary ray. On the other hand, if the e-vector oscillates perpendicularly to a principle plane, then it sees an index of refraction , and the ray direction and wave-normal di- rection are parallel. This beam of light is known as the ordinary ray. If unpolarized light enters a birefringent crystal, there will be e-vector components parallel and perpendicular to a principle plane, in general. The electric field components perpendicular to a principle plane (ordinary ray) will pass straight through, while the electric field components parallel to a principle plane Fig. 3. Wire-grid dichroic . A grid of conductive wires absorbs one e-vector component of incident radiation. Figure was adapted from [44]. (extraordinary ray) is displaced by a distance that depends on the thickness of the crystal and on the angle of refraction [Fig. (2b)]. allel to the wires (in ), and the other be perpendicular (in ). Hence, an image appears to be doubled when seen through a The e-vector parallel to the conductors will excite mobile elec- calcite crystal (although the intensity of each image is only half trons to oscillate along the length of the conductor. The excited of the original for unpolarized incident light). The double im- electrons will lose this imparted energy by colliding with lattice ages are completely linearly polarized, and their e-vectors are atoms, transferring their energy into kinetic energy (heat) in the orthogonal relative to one another. If we look at the double im- lattice. This energy transfer (from electromagnetic to kinetic) ages through another calcite crystal we will still see the double causes the e-vector of the transmitted wave along the direction image, but the relative intensity between these images can be of the conductors to be attenuated, resulting in mostly linearly modulated by rotating the second crystal relative to the first (just polarized light orthogonal to the orientation of the wire grid on like Huygens discovered in the late 17th century). the other side. Polaroid films, or H-sheets, like the sheet in- B. Dichroism or Polarization by Absorbtion vented by Land [45], [46] use this same principle on a much When an unpolarized light ray passes through a dichroic po- smaller scale. Dense and thin arrays of “micro-wires,” or larizer, it emerges as predominantly linearly polarized light at dichromophores, composed of conducting crystals, molecules, about half the intensity as the original ray. While a birefringent or small-thin strips of metal [47], are aligned along a common crystal segregates orthogonal e-vector components into two di- axis either magnetically, electrostatically, or mechanically, verging rays inside the crystal, a dichroic polarizer selectively depending on the materials used and the technique applied absorbs one of the e-vector components, while allowing the or- [45], [44], [48]. These dichromophores absorb radiation at thogonal component to pass through unaffected. optical frequencies, converting optical energy along the axis Consider a macroscopic example of a dichroic polarizer: the of absorption into heat. The alignment process affects all wire-grid polarizer (Fig. 3). It consists of an array of parallel the dichromophores simultaneously, creating a global axis of conductors (wires). We resolve electromagnetic radiation inci- absorption throughout the material. dent upon the wire-grid polarizer into orthogonal components, In this work, we will deal mostly with polymer based as we are used to doing by now: let one component be par- dichroic films, and we will show how to pattern a polymer ANDREOU AND KALAYJIAN: POLARIZATION IMAGING: PRINCIPLES AND INTEGRATED POLARIMETERS 569 dichroic polarizer so that it exhibits a high resolution spatial splitter is described. In this system, light from a scene passes pattern of polarization. through a polarizing . The beam splitter separates polarization components by reflection one orientation of polar- IV. IMAGING POLARIMETERS ization (whose e-vector is parallel to the plane of incidence) will be reflected and transmitted in a certain proportion different In the previous section, we described the quality of light from the orthogonal orientation. The intensity recorded by the known as polarization and discussed how light becomes polar- CCD cameras is a separable mixture of two orthogonal polar- ized. There have been many inventions that have been used to ization states in the image. The intensity images are piped into analyze the polarization of a beam of light, like the Nicol , a computer, where the orthogonal components are unmixed, and Wollastan prism, Glan-Foucault polarizer, and the H-sheet. At where polarization contrast is computed. Camera registration is least one of these polarizers can be found in practically any also an issue for this system, which is mounted on an optical lab in the world. The H-sheet has also become popular bench for laboratory use. outside the laboratory, being used in photography, liquid crystal Andreou and Wolff also used liquid crystal rotators in a polar- displays, and in sunglasses. It was the H-sheet that allowed ization camera [36]. A liquid crystal rotator is an electro-optic Karl von Frisch to first examine the effect of polarization on device capable of rotating the angle of linear polarization of the orienting behavior of the honey-bee in 1948 [1]. However, light passing through it without changing the intensity. Liquid the ability to characterize polarization over a whole scene had crystals, which are anisotropic, use birefringence to change to wait until the development of electronic imaging technology. the orientation of linear polarization; however, light is always In this section, we will look at polarization imagers—devices guided along the optic axis and thus never separates into two that measure the Stokes vector (4) for each point in an image. beams. Instead, the birefringence of the liquid crystal redirects All imaging polarimeters share a common photon detection the e-vector as it travels through the liquid crystal matrix mechanism: electron-hole pair generation by photon absorption [50]. One liquid crystal rotator can redirect the orientation in a semiconductor. Photon absorption in a semiconductor de- of linear polarization up to 90 . In this system, two rotators, pends only on the energy (frequency) of the photon; it does not each set to rotate by 45 , were used in tandem, permitting four depend on the polarization of the photon. Therefore, the chal- combinations of rotation: 0 ,45,90, and 135 . The liquid lenge of detecting different forms of polarization using a semi- crystal based polarization camera used only the 0 ,45, and conductor based imaging device becomes that of transforming 90 directions. polarization information into intensity, a quality of light which In the liquid crystal polarimeter, one CCD camera, with a a semiconductor detector is capable of distinguishing. As we lens-mounted linearly polarizing filter, is placed behind the two discussed in the previous section, there are four basic physical liquid crystal rotators. The linear polarizer remains fixed at one processes that can transform polarization into intensity—reflec- orientation relative to the camera and the liquid crystal rota- tion, scattering, dichroism, and birefringence. Of these four two tors. The rotators are switched on and off in synchrony with the of them, dichroism and birefringence are the ones that are em- CCD camera frame rate. Consequently, each consecutive frame ployed in the integrated polarization sensors that are described of intensity recorded by the CCD camera corresponds to a dif- in this paper. ferent rotated image. Thus, within three image frames, the liquid crystal polarization camera samples the scene at three different A. Imaging Polarimeter Designs orientations of linear polarization. In effect, instead of rotating In this section, we will briefly review work in the design of a linear polarizer in front of the camera lens to capture different imaging polarimeters. This is by no means a complete review of orientations of polarization in a scene, the liquid crystals ro- the literature; a more elaborate review is presented in [35]. tate the scene in front of the camera. Three samples of linear Although it violates the definition of an imaging polarimeter polarization are digitized and piped to a computer. Three sam- as stated earlier, we describe here the system used ples are enough to completely characterize the linearly polar- by Roger, et al., because it exemplifies the process required ized component of light in a scene; in Stokes vector parlance, for creating a polarization image [23]. Roger’s system, which the liquid crystal polarization camera can detect , , and . was used to take pictures of the Earth and its atmosphere from This system cannot discern . Capture frame the U.S. Space Shuttle, uses two photographic Hasselblad cam- rates are limited to a few frames per second by the liquid crystal eras to capture images through crossed linearly polarizing fil- rotators, which have switching speeds on the order of 10 ms. ters. The cameras were registered so that each saw the same A compact polarization camera was assembled using a cam- field of view. The cameras simultaneously recorded an expo- corder. The recorded intensity data from the polarization cam- sure through the linear polarizers on black and white film. The corder was downloaded to a computer that extracted polariza- film was subsequently developed and digitized. A polarization tion from the video data. contrast image was extracted from the digitized snapshots by a Another system that uses electrooptic devices to modulate an computer. This process fell victim to a number of problems: reg- image is described in [51]. In this work, all four Stokes param- istration of the images, non-linearity in the film response, and eters are extracted. The system uses piezoelastic modulators, calibration of the digitized images. which are crystalline materials that exhibit birefringence under Wolff and Andreou have designed a number of imaging po- electrically induced mechanical stress. The system uses a com- larimeters for computer vision [35]. In [49], an imaging po- plex optical bench setup that includes three CCD cameras, two larimeter that uses two CCD cameras and a polarizing beam piezoelastic modulators, two beam splitters, and one linear po- 570 IEEE SENSORS JOURNAL, VOL. 2, NO. 6, DECEMBER 2002

Fig. 4. (a) Photomicrograph of 1-D polarization contrast retina. The photograph was taken using the microscope’s built-in linear polarizer oriented parallel to the ƒ-phase region of the two-phase dichroic film. (b) Output from the polarization contrast circuit was measured while a linear polarizer was rotated in front of the polarization retina. larizer. The CCD cameras and piezoelastic modulators, which film consisting of two dichroic regions that have orthogonal are driven by a 50 kHz sinusoidal driving signal, need to be care- absorption axes. This dichroic film was obtained through a fully synchronized. Captured intensity images are again piped commercial fabricator of Polaroid film based advertisement to a computer for computation of a polarization image. animations—Frank Woolley & Company. We will refer to The imaging polarimeters described above are complicated these regions of orthogonal polarization as the -phase and the electro-optic systems, but they have in common three basic -phase regions. components: optics that transform polarization into intensity, The and regions of the dichroic film are aligned with an intensity detector/imaging device, and a computer that the photodiode array such that one photodiode in each pair lies calculates Stokes parameters. Each of these components tends underneath the region of the dichroic film and the other pho- to be bulky and power hungry. Furthermore, the instruments todiode lies underneath the region. Thus, the photogenerated tend to be one-of-a-kind thus not being as widely available as currents of each photodiode in a pair represents the intensity video cameras. of the incident light whose e-vector component oscillates par- The article by Thompson et al [52] was the first report of allel to the and axes, respectively. As the linear array is se- a truly integrated polarization imager (reduced Stokes vector quentially scanned, the photocurrents from one pair of selected ( , ) pixels). A system design was presented but due to manu- photodiodes are routed through multiplexing circuitry to the pe- facturing difficulties in obtaining a large are CCD based imager riphery of the chip for processing. in non optimized two polysilicon CMOS process the imager was At the periphery of the chip, the photocurrents representing never functional and remained as a paper design. the intensity of and polarized light are routed to a translinear Motivated by our earlier successes as well as the difficulties circuit that computes polarization contrast, defined as that our colleagues had at the Applied Physics Laboratory [52], we proceeded to explore the possibility of a CMOS based, truly Polarization Contrast (7) integrated polarization imager [37]–[40], [53]. In contrast to the imaging polarimeters discussed in the previous section, the Note that polarization contrast is not the same, in general, as highly integrated imaging polarimeter, or polarization retinae, degree of polarization. In other words, may not rep- incorporate all three components of polarization imaging on the resent the intensity of polarized light: e.g., completely linearly focal plane of a CMOS imager: 1) optics that detect orthogonal polarized light which falls at a 45 angle relative to the and e-vector components, 2) an imaging array of photodetectors, orientations would have a polarization contrast of 0. Polariza- and 3) circuitry that computes a useful measure of polarization tion contrast measures a mixture of degree of polarization and for each point in an image. We named these class of imagers po- orientation of linear polarization. larization retinae because they were partly inspired by the eye A photomicrograph of the 1-D polarization contrast retina is architectures of living creatures. shown in Fig. 4. The internal microscope polarizer was adjusted so that the -phase of the dichroic film would be visible. The B. Dichroic Film-Based Polarization Contrast Imager -phase is opaque due to the canceling effect of two orthogonal Our first polarization retina used a dichroic polymer film linear polarizers. Photodiodes and scanning circuitry are visible mounted on the focal plane of a one–dimensional (1-D) CMOS underneath the -phase region of the focal plane polarizer. As imager [37], [53]. The CMOS imager consists of a 1-D array the chip is scanned, two selected photodiodes are selected, one of 29 photodiode pairs integrated in a 2 m CMOS process. under each polarizer region. All chip fabrication was done through the MOSIS service. On The chip was tested by rotating a linear polarizer in front of top of the linear array of photodiodes we mounted a dichroic 1-D polarization contrast retina while measuring the polariza- ANDREOU AND KALAYJIAN: POLARIZATION IMAGING: PRINCIPLES AND INTEGRATED POLARIMETERS 571

a custom fabricated CMOS imager and focal plane micropolar- izer [39], [40]. The two-phase micropolarizer, was aligned and attached to a custom designed CMOS imager. Concentrating on the intensity images on the left and right of Fig. 6, bright regions due to specular reflections from these light sources are visible at many points on the object. Two prominent extended regions of specular reflection in particular are pointed out. In the intensity images, these regions of specular reflection are nearly identical. In fact, these bright spots look identical when observed with the naked eye. However, we know that light will acquire linear polarization when reflected from a dielectric surface. The amount of polarization will depend on the angle of reflection and the material properties of the reflecting surface. Fig. 5. Cross-sectional view of a high resolution polymer based focal-plane Indeed, the polarization contrast image (center) reveals that polarizing filter and CMOS imager. these specular reflections are distinct in their polarizations. The bright region on the right side of the object is highly polarized tion contrast output at 5 intervals. The theoretical response for along the orientation, being reflected at a larger angle with re- the system is given by [40] spect to the surface normal than the more central bright region. Most of the visible features in the polarization contrast image (8) appear either in regions of high specular reflection, or at other which is shown to model the experimental data well. surface features of the object, such as occluding contours. One A 1-D polarization imager is of limited usefulness for com- notices first that most of the image is gray, whereas the and puter vision or for polarization-based scene analysis. However, images show a wider dynamic range in intensities. The contrast we are confronted with a problem when we consider extending computation compresses the dynamic range of the image by nor- the above 1-D polarization imager that uses dichroic films into malizing each point the total intensity, , of polarized and a high resolution two-dimensional (2-D) polarization imager: unpolarized light at that point. Shadows that are apparent in the The bulk processing of dichroic films (i.e., stretching the film intensity images are invisible in the polarization contrast image to align PVA molecules) precludes fine structure in the orienta- as they contain no polarization information. tion of linear polarization. Today’s CMOS imagers have pixel Unlike the imaging polarimeters described in the previous sizes on the order of 5 m. How do we get a dichroic film with section the polarization contrast retina is compact and efficient, pixel-sized features of linear polarization? requiring only a 50 mW of power (under room lighting con- While it may be possible to grow or form aligned PVA ditions), an external clocking signal for video scanning, and a molecules in a way that exhibits a spatially varying absorption lens. Its power and space requirements make it more suitable for axis at the micron scale, we approached the problem from an mobile and remote sensing applications than other imaging po- alternative perspective. Instead of changing the orientation larimeters. This high resolution 2-D polarization contrast retina of the dichromophores from one point in the dichroic film is the first to successfully integrate on the focal plane the three to the next, we develop a processing technique that allows functions required of an imaging polarimeter: re-coding polar- us to spatially modulate the presence of dichromophores in a ization into intensity, light transduction, and computation of a dichroic PVA film [38]. With this capability, we can assemble polarization metric. a polarizing filter comprising multiple dichroic films that have The 2-D polarization contrast retina is also a fully cur- been processed in such a way as to exhibit a spatial pattern of rent-mode CMOS imager, where transduced intensity is linear polarization. communicated and processed throughout the chip in the current Consider one such film for a 2-D polarization contrast retina, domain. Even though this polarization retina was capable of a schematic of which is shown in Fig. 5. Two dichroic polymer producing dynamic images, its maximum speed of operation based polarizers, whose absorption axes are orthogonal to each was only a few frames per second as it was never optimized for other, are processed so that each film has an alternating pattern speed and its output is only a reduced Stokes vector ( , ). of dichroic and transparent stripes. The patterns are complemen- It is however a proof of concept that a three state polarization tary; that is, when the films are properly aligned the dichroic imager capable of producing a complete description of light at stripes from one film interdigitated with the dichroic stripes each pixel is possible. from the other film. While the stripes of dichroic PVA are par- allel between the two films, their absorption axes are orthog- C. Birefringent Crystal-Based Polarization Contrast Imager onal. The end result is a focal-plane micropolarizer: a polarizing The reduced vector polarization imager is realized using a filter with micron-scale regions (stripes) of linear polarization. custom designed CMOS imager and interface electronics. This These patterns are lithographically defined to match the scale of is so because the geometrical arrangement that forms the polar- a custom designed high resolution CMOS imager. ization metapixel can be accommodated easily within the pixel Utilizing the micropolarizer fabrication techniques (de- layout of custom designed active pixel sensor arrays. The po- scribed in detail in Appendix A), we constructed a 2-D larization dependent beam splitting is achieved using a birefrin- high-resolution integrated polarization retina, which comprises gence crystal mounted at the focal plane. 572 IEEE SENSORS JOURNAL, VOL. 2, NO. 6, DECEMBER 2002

Fig. 6. Video images from the integrated 2-D polarization contrast retina using polymer-based focal-plane micropolarizer. In the picture, a black plastic object is illuminated from two angles by two light sources. The center image represents polarization contrast, @s s Aa@s C s A; while the left and right images represent respectively s and s : the intensity of light passing through the orthogonal linearly polarization filters. In the s and s intensity images, bright regions of specular reflection are indistinguishable from one another. However, in the polarization contrast image, these specular reflections are shown to be different in their polarization because of different angles of reflections off of the smooth dielectric surface.

Fig. 7. Cross-sectional view of a high resolution birefringent crystal based focal-plane polarizing filter and CMOS imager. The birefringent crystal separates orthogonal components of linearly polarized light. The extraordinary ray is refracted from the orthogonally polarized ordinary ray by a fixed angle  (% T for TiO ). The crystal is lapped to a thickness which exactly displaces the orthogonal components a distance equal to one phototransistor width. An aluminum shield is deposited on the surface of the crystal to prevent mixing of ordinary and extraordinary components.

In an earlier section, we discussed how a birefringent crystal, The crystal is specially prepared before it is mounted on the such as calcite, refracts incident unpolarized light rays at two focal plane. First, we must ensure that the ordinary and extraor- different angles, depending on the e-vector orientation of the dinary components exit the crystal in the right location, i.e. over linearly polarized components. One ray, called the ordinary ray, adjacent photodetector sites. The ordinary ray, obeying Snell’s refracts like a normal beam of light through a regular dielectric law, will pass straight through the crystal since it enters at a medium. The other ray behaves quite extraordinarily in that the normal angle of incidence. The extraordinary ray, however, will angle and direction of refraction are different from the ordinary break the law of refraction and be bent at an angle determined by ray, hence it is known as the extraordinary ray. Upon exit, the the orientation of the optical axis relative to the surface normal. ordinary and extraordinary ray are completely orthogonally lin- Rutile has a high degree of birefringence, which means the dif- early polarized. ference between the ordinary and extraordinary indices of re- If the incident radiation is itself polarized, the relative intensi- fraction is large. For rutile, and . The ties of the ordinary and extraordinary rays will reflect the orien- refraction angle between the two orthogonally polarized rays is tation of linear polarization of the incident radiation with respect given by [42] to the principle plane. In other words, if the incident light is com- pletely linearly polarized parallel to a principle plane, then all the radiation through the crystal will be extraordinary; the con- (9) verse is also true for light perpendicularly polarized with respect to the principle plane. Therefore, using the property of birefrin- where is the refraction angle, and is the angle between gence, we can measure the composition of an incident ray in the optical axis and the surface normal. The refraction angle terms of orthogonal linearly polarized components. reaches a maximum value of 6 when is 45 . In order to max- This is how we designed a polarization contrast imager using imize separation between the ordinary and extraordinary rays, a birefringent crystal based focal-plane micropolarizer. Fig. 7 we cut the crystal so that the optical axis is oriented at 45 to shows a cross-section of the device. the surface normal. ANDREOU AND KALAYJIAN: POLARIZATION IMAGING: PRINCIPLES AND INTEGRATED POLARIMETERS 573

Having chosen the orientation of the surface normal of the rutile crystal, we next have to choose a thickness that will ap- propriately displace the extraordinary ray. A simple geometric calculation shows that if the extraordinary ray is refracted 6 from the surface normal, it will be displaced by 12.5 m (a typ- ical distance between adjacent photodetector cites in an imager) after having passed through a thickness of 118 m of rutile. Doing lithography on a 118 m thin material is challenging but is doable if the samples are appropriately mounted on silicon wafers. The final step before mounting the micropolarizer on the focal plane of the CMOS imager is to deposit a Ronchi ruling on the top surface of the rutile crystal. A Ronchi ruling is a grating of opaque lines whose width and spacing are equal to the width of a Fig. 8. Polarization contrast is computed one pixel at a time by this translinear photodetector site. The opaque grating is formed using the same circuit. The circuit uses a bipolar transistor translinear loop from e to e .In deposition and lithographic techniques that are used to deposit this case, the translinear loop is a virtual one: the emitters of I and R are kept at the same potential, † , using feedback through the transconductance a metal interconnect layer in a CMOS fabrication processes. amplifiers. The rutile crystal is mounted onto a glass wafer using a UV curing epoxy. This allows the birefringent crystal to be treated as if it were a silicon wafer going through a CMOS fabrica- contrast imager produces sharper and more uniform images as tion process. Aluminum is evaporated onto the top surface of the compared to the dichroic polymer based one. This is because crystal in an evaporation chamber. Photoresist is spun onto the the processing steps for the birefringent crystal based microp- aluminum, and then exposed through a mask, which is itself a olarizers are fewer and less likely to produce variation across Ronchi ruling. Then, the photoresist is developed and hardened the surface of the micropolarizer. Alignment is also simpler for in a post-bake. The rutile crystal is removed from the glass wafer the crystal micropolarizer imager: only one alignment step is re- (the epoxy is dissolved by acetone), and is then immersed in an quired (between the micropolarizer and CMOS imager), while aluminum etchant. Etching continues until all the exposed alu- the polymer micropolarizers requires three alignment steps (be- minum is dissolved from the crystal surface; the hardened pho- tween the three layers of the micropolarizer filter, and between toresist protects the aluminum underneath it from dissolution. the micropolarizer and CMOS imager.) The final result is a pattern of opaque aluminum lines on The birefringent crystal micropolarizer is also more efficient the top surface of the rutile crystal. The cross-sectional view in separating orthogonal linearly polarized components. Unpo- in Fig. 7 shows the aluminum strips covering every other pho- larized light entering a birefringent crystal will be divided into todetector. The photodetectors underneath the Ronchi ruling are two beams, each beam comprising exactly 50% of the incident protected from direct incident light (ordinary radiation); they re- radiation. The same does not hold for dichroic polymer film ceive only radiation that is refracted from an adjacent photode- polarizers. Unpolarized light passing through a dichroic film tector site (extraordinary radiation). should ideally emerge at 50% of the incident intensity, since The birefringent crystal micropolarizer is then aligned to the unpolarized light oscillates in parallel to the absorption axis of CMOS imager and affixed with UV curing epoxy. Alignment is the dichroic film statistically half the time. However, the molec- performed under a microscope with an internal linear polarizer. ular alignment process used in the making of polymer dichroic When the microscopes polarizer is oriented in one direction, films is not perfect, and some of the incident radiation parallel the photodetectors directly visible through the Ronchi ruling to the transmission axis is also absorbed. The dichroic films appears. If the internal polarizer is rotated 90 , the refracting that will be used for the polymer focal-plane micropolarizer property of the rutile crystal allows the viewer to see the pho- (Polaroid film HN-38) have a transmission efficiency of 38%. todetectors underneath the opaque grating. Consequently, the birefringent crystal micropolarizer preserves The readout interface to the sensor controls the random ac- more incident radiation, and is therefore a more efficient polar- cess imager and include computational readout circuits to do the izer. analog arithmetic and compute polarization contrast. Polariza- However, image non-uniformities inherent in the CMOS im- tion contrast is simply the ratio of to . Arithmetic operations ager itself and due to the micropolarizer fabrication can be com- on the chip are performed using current-mode Translinear cir- pensated electronically on-chip and off-chip in the software and cuits as done with our previous work on polarization contrast are not expected to degrade the performance of the three Stokes retinae [37]. vector polarization imager. A polarization contrast arithmetic circuit is shown in Fig. 8; currents and are the inputs generated from readout APPENDIX A voltages and is the output. FABRICATION OF THE POLYMER MICROPOLARIZER ARRAY H-sheets or Polaroid films are formed by dissolving V. D ISCUSSION solution into a sheet of PVA, usually in the presence of boric If we compare the performance of the two type of micropolar- acid [54]. If the iodine-doped PVA is heated and stretched, izers, we expect that the birefringent crystal based polarization the long polymer chains are axially coordinated along the 574 IEEE SENSORS JOURNAL, VOL. 2, NO. 6, DECEMBER 2002 stretching direction. Stretching PVA causes an abundance of not completely flat after the acetate etch, it must be polyiodine complexes of or to form parallel to the remounted on the glass wafer. First one side (e.g., polymer molecules [55], [56]. These conducting complexes north side of the square specimen) is securely taped form a light absorbing axis in the film along the stretching to the glass wafer; then, the other side is stretched direction, imparting dichroism to the polymer sheet. Since back, so that the film becomes flat, and then secured making an H-sheet involves stretching the entire bulk of the to the glass wafer. East and west sides of the square PVA film in one direction, the entire sheet exhibits dichroism Polaroid film are also stretched flat in this way. Once along one axis—the stretching axis. This method of production an area of PVA is exposed, a layer of photoresist is of polymer dichroic films is not conducive to making a focal spun onto the specimen. The OCG 820 photoresist plane polarization filter for high resolution polarimetry. used in these experiments was spun at 4500 rpm However, by lithography and etching techniques similar to for 60 s. This layer of photoresist is pre-baked in a those found in the CMOS fabrication process, it is possible to convection oven at 100 C for 5 min. dissolve, or undope, selectively the iodine dichromophores from Step 3) Expose Photoresist through mask: A UV opaque the PVA substrate, thereby locally destroying the dichroism of mask is used as a template to define which areas the linear polarizer. A pattern of dichroic and transparent re- of PVA are to be cleared of dichromophores (un- gions, defined by a lithographic mask, can be attained on the doped). Using a mask aligner, the photoresist layer same PVA substrate, which enables the creation of high resolu- just deposited on the specimen is exposed to UV tion polarization filter that has been employed in the 2-D polar- light through a chromium photolithographic mask ization contrast retina [38], [39]. A similar process can be used for 15 s at an intensity 8 mW/cm . The procedure to selectively dope iodine into clean PVA material [57]. employs a single dichroic film that is tripled-exposed The primary challenge in working with Polaroid films—we through the same lithographic mask, instead of ex- use commercially available type HN-38 Polaroid—is its me- posing two dichroic films individually. In this triple- chanical flexibility. Keeping the Polaroid film flat throughout exposure technique, two thirds of the photoresist- the photolithographic process is essential. In order to keep the covered dichroic film is completely shielded with a film flat, it is cut square and taped on four sides to a glass UV opaque cover (aluminum foil was used); and one wafer (about the same dimensions as a silicon wafer). Heat re- third is exposed through the photolithographic mask sistant PTFE based tape, which was found satisfactorily im- that will define the pattern of dichroism in the lin- mune to both chemical and thermal exposure, can be used to early polarizing film. Next, the other third is shielded mount the Polaroid film on the glass wafer. Using a silicon with the UV opaque cover. The photolithographic wafer-sized mount has an additional advantage: the flexible film mask is rotated 90 and moved over one region of is easily interfaced with the necessary silicon processing equip- the dichroic film, which has remained unexposed to ment, namely the spinner and mask aligner, that are designed UV. This is exposed through the photolithographic specifically for silicon wafer processing. mask. The procedure is repeated at a 45 . The end Step 1) Acetone Etch: H-sheets consist of an iodine-doped result is that the three spatially distinct are of the PVA core sandwiched between layers of cellulose same dichroic film are exposed through different ori- acetate or polyethylene, which provide structural entations of the same lithographic mask. Processing stability and prevent water absorption and swelling continues on this one dichroic film. At the end of the of the PVA. Consequently, in order to perform process, each region will have undergone an iden- lithography on the PVA layer, the acetate must be tical set of fabrication steps, and therefore be very etched away. Acetone is used as the etching reagent closely matched. After UV exposure is complete for in this step. Since the film is taped to a glass mount all three regions of the dichroic film, it is post-baked using a chemical resistant tape, only a 3–4 cm in a convection oven at 100 C for 5 min to anneal window (on one side of the film) is exposed to the areas of unexposed photoresist. acetone etch. The acetate layer is etched by sub- Step 4) Over-development: The UV exposed film is im- merging the entire film, mounted on the glass wafer, mersed in photoresist developer (160 ml KTI: 640 ml into an acetone bath. After 3 min in the acetone DI). The developer dissolves areas where photoresist bath, the exposed cellulose acetate becomes soft and has been exposed to the UV light through the litho- begins to loosen from the PVA. Vigorous rinsing graphic mask. In a normal silicon lithographic pro- with de-ionized water (DI) at this point washes off cedure, a wafer stays in the developer solution until much of the softened acetate. The softened acetone the exposed photoresist is fully dissolved—a process solidifies under DI in a process called extrusion, that takes about 30 s for the combination of pho- and is easily washed off the PVA. An additional 3–4 toresist and developer used in this procedure. How- min in acetone followed by another DI rinse clears ever, this process requires that the film remain in the away any remaining cellulose acetate. At this point, developer solution longer than normally required. the inner PVA layer, which is immune to acetone, Once the UV exposed areas of photoresist are com- is exposed. pletely dissolved, the iodine-doped PVA underneath Step 2) Spin on Photoresist: Occasionally, the hydrophilic is exposed to the basic developer. Iodine is dissolved PVA will curl after exposure to DI. If the film is in these regions by the developer, thereby destroying ANDREOU AND KALAYJIAN: POLARIZATION IMAGING: PRINCIPLES AND INTEGRATED POLARIMETERS 575

the conductive complexes in the PVAresponsible for [15] C. Hawryshyn, “Ultraviolet-polarization vision: its role in salmon navi- dichroism. At room temperature, exposing the pho- gation,” in Proc. SPIE, vol. 3120, 1998, pp. 2–10. [16] D. Cameron and E. Pugh Jr., “Double cones as a basis for a new type of toresist covered PVA to the developer for roughly 2 polarization vision in vertebrates,” Nature, vol. 353, pp. 161–164, 1991. min is sufficient to break up the exposed photoresist [17] M. Rowe, N. Engheta, S. Engheta Jr., and E. Pugh, Jr., “Graded-index of and completely dissolve the dichromophores from a fish double cone exhibits differential polarization sensitivity,” J. Opt. Soc. Amer. A, vol. 11, no. 1, pp. 55–70, 1994. out of the PVA substrate. [18] K. Adler and D. Taylor, “Extraocular perception of polarized light by Step 5) Alignment and bonding: Once the dichroic film has orienting salamanders,” J. Comput. Physiol., vol. 87, pp. 203–212, 1973. been developed, the patterned areas are cut out from [19] K. Adler and J. Phillips, “Orientation in a desert lizard: time compen- sated compass movement and polarotaxis,” J. Comput. Physiol., vol. the original dichroic film. The first of the three re- 156, pp. 547–552, 1985. gions of the patterned film is re-attached to the glass [20] N. Shashar, P. Rutledge, and T. Cronin, “Polarization vision in cuttle- wafer and mounted in the mask aligner. The second fish—an concealed communication channel?,” J. Exper. Biol., vol. 199, pp. 2077–2084, 1996. region is attached to a blank photolithographic mask [21] K. Coulson, Polarization and Intensity of Light in the Atmosphere, 1st and placed into the mask-holder of the mask-aligner. ed. Hampton, VA: A. Deepak, 1988. Now, using the mask aligner as it was intended, the [22] H. Holtzclaw, W. Robinson, and W. Nebergall, General Chemistry, 7th ed. Boston, MA: Heath, 1984. two regions are aligned to each other so that the po- [23] J. Roger, R. Santer, M. Herman, and J. Deuz, “Polarization of the solar larizing squares of the first region lie on top of the light scattered by the earth-atmosphere system as observed from the non-polarizing squares of the second region. When shuttle,” Remote Sens. Environ., vol. 48, pp. 275–290, 1994. [24] E. Vogt and J. Henoux, “Polarimetric study of solar flares,” Solar Phys., good alignment is achieved, one piece of the pat- vol. 164, pp. 345–359, 1996. terned film is lightly coated with a UV-curing epoxy. [25] L. Wolff and T. Boult, “Constraining object features using a polarization Then, the other piece is brought into contact. When reflectance model,” IEEE Trans. Pattern Anal. Machine Intell., vol. 13, pp. 635–657, July 1991. satisfactory alignment is achieved, the UV lamp of [26] H. Chen and L. Wolff, “Polarization phase-based method for material the mask aligner is activated to expose for 5 min the classification and object recognition in computer vision,” Int. J. Comput. UV-curing epoxy. The process is repeated for the Vis., vol. 28, no. 1, pp. 128–135, 1996. [27] T. Jenkins, “Multiple-angle-of-incidence elipsometry,” Phys. D: Appl. third pattern, and the three layer sandwich assembly Phys., vol. 32, pp. 45–56, 1999. is aligned and mounted on the CMOS imager chip [28] L. Denes, M. Gottlieb, B. Kaminsky, and D. Huber, “Spectro-polari- using the procedure used for the birefringent beam metric imaging for object recognition,” in Proc. SPIE, vol. 3240, 1998, pp. 8–18. splitter. [29] L. Balfour, Y. Bushlin, and N. Brandman, “Polarization signatures of man made objects in the infrared spectral band,” in Proc. SPIE, vol. 3375, 1999, pp. 48–53. REFERENCES [30] B. Barbour, M. Jones, H. Barnes, and C. Lewis, “Passive polarization [1] K. von Frisch (translated by Leigh E. Chadwick), The Dance Language sensors: a new technology for mine detection,” in Proc. SPIE, vol. 3392, and Orientation in Bees. Oxford, U.K.: Oxford Univ. Press, 1967. 1999, pp. 96–103. [2] W. Edrich and O. V. Helversen, “Polarized light orientation in honey [31] S. Nayar, X. Fang, and T. Boult, “Separation of reflection components bees: is time a component of sampling?,” Biol. Cybern., vol. 56, pp. using color and polarization,” Int. J. Comput. Vis., vol. 21, no. 3, pp. 89–96, 1987. 163–186, 1997. [3] K. Bartsch, “Polarization-sensitive photoreceptors of different spectral [32] S. Lin and S. Lee, “Detection of specularity using stereo in color and types in the compound eye of waterstriders,” Naturwissenshaften, vol. polarization space,” Comput. Vis. Image Understand., vol. 65, no. 2, pp. 82, pp. 292–293, 1995. 336–346, 1997. [4] , “Reflection-polarization patterns at flat water surfaces and their [33] D. Hunter, S. Patel, and D. Guyton, “Automated detection of foveal fix- relevance for insect polarization vision,” J. Theor. Biol., vol. 175, pp. ation by use of retinal birefringence scanning,” Appl. Opt., vol. 38, no. 27–35, 1995. 7, pp. 1273–1279, 1999. [5] G. Horv and D. Varj, “Underwater refraction-polarization patterns of [34] A. Vickers, “The polarizing microscope in organic chemistry and bi- skylight perceived by aquatic animals through Nell’s window of the flat ology,” Research, vol. 9, pp. 67–78, 1956. water surface,” Vis. Res., vol. 35, no. 12, pp. 1651–1666, 1995. [35] L. Wolff and A. Andreou, “Polarization camera sensors,” Image Vision [6] , “Spectral regions in which aquatic insects see reflected polarized Comput., vol. 13, pp. 497–510, Aug. 1995. light,” J. Comput. Physiol. A, vol. 177, pp. 439–448, 1995. [36] L. Wolff, T. Mancini, P. Pouliquen, and A. Andreou, “Liquid crystal po- [7] R. Schwind, “Polarization vision in water insects and insects living on a larization camera,” IEEE Trans. Robot. Automat., vol. 13, pp. 195–203, moist substrate,” J. Comput. Physiol. A, vol. 169, pp. 531–540, 1991. Apr. 1997. [8] N. Marshall, M. Land, C. King, and T. Cronin, “The compound eyes [37] Z. K. Kalayjian, A. Andreou, and L. Wolff, “ID polarization contrast of mantis shrimp (crustacea, hoplocarida, stomatopoda) compound eye retina,” IEE Electron. Lett., vol. 33, pp. 38–40, Jan. 1997. structure: the detection of polarized light,” Phil. Trans. R. Soc. Lond. B, [38] Z. Kalayjian, A. G. Andreou, L. B. Wolff, and N. Sheppard, “A polar- vol. 334, pp. 33–56, 1991. ization sontrast retina that uses patterned iodine-doped PVA film,” in [9] M. Shashar and T. Cronin, “Polarization contrast vision in octopus,” J. Proc. 22nd Eur.Solid-State Circuits Conf., Neuchatel, Switzerland, Sept. Exper. Biol., vol. 199, pp. 999–1004, 1996. 1996, pp. 308–311. [10] R. Glantz, “Polarization sensitivity in crayfishlamina monopolar neu- [39] Z. Kalayjian and A. Andreou, “Integrated high resolution focal-plane rons,” J. Comput. Physiol. A, vol. 178, pp. 413–425, 1996. polarization imager,” in Proc. ISA EXPO, Austin, TX, Nov. 1998. [11] T. Labhart, E. Meyer, and L. Schenker, “Specialized ommatidia for po- [40] Z. Kalayjian and A. Andreou, “Silicon retina for polarization contrast vi- larization vision in the compound eye of cockchafers,” Cell Tissue Res., sion,” in Proc. Int. Conf. Neural Networks, Washington, DC, July 1999. vol. 268, pp. 419–429, 1992. [41] Z. Kalayjian, “Optoelectronic Vision and Image Processing,” Ph.D. dis- [12] A. Philipsborn and T. Labhart, “A behavioral study of polarization vision sertation, Johns Hopkins Univ., Baltimore, MD, 2000, unpublished. in the fly,” J. Comput. Physiol. A, vol. 167, pp. 737–743, 1990. [42] M. Born and E. Wolf, Principles in Optics, 3rd ed. New York: Perg- [13] D. Couglin and C. Hawryshyn, “A cellular basis for polarized-light vi- amon, 1965. sion in rainbow trout,” J. Comput. Physiol. A, vol. 176, pp. 261–272, [43] P. Feofilov, The Physical Basis of Polarized Emission, 1st ed: Consul- 1995. tants Bureau, 1961. [14] D. Parkyn and C. Hawryshyn, “Polarized-light sensitivity in rainbow [44] E. Hecht, Optics, 2nd ed. Reading, MA: Addison-Wesley, 1987. trout: characterization from multi-unit responses in the optic nerve,” J. [45] E. Land and C. West, “Dichrosim and dichroic polarizers,” Colloid Comput. Physiol. A, vol. 172, pp. 493–500, 1993. Chem., vol. 6, no. 12, pp. 160–190, 1946. 576 IEEE SENSORS JOURNAL, VOL. 2, NO. 6, DECEMBER 2002

[46] E. Land, “Dichroism and dichroic polarizers,” J. Opt. Soc. Amer., vol. Andreas G. Andreou received the Ph.D. degree 41, no. 12, p. 957, 1951. in electrical engineering and computer science in [47] G. Nordin, J. Meier, P. Deguzman, and M. Jones, “Micropolarizer array 1986 from The Johns Hopkins University (JHU), for infrared imaging polarimetry,” J. Opt. Soc. Amer. A, vol. 16, no. 5, Baltimore, MD. pp. 1168–1174, 1999. Between 1986 and 1989, he held Post-Doctoral [48] R. Slocum, “Evaporative thin metal films as polarizers,” in Proc Fellow and Associate Research Scientist positions SPIE—Polarizers Applicat., vol. 307, 1981, pp. 25–30. in the Electrical and Computer Engineering De- [49] L. Wolff, “Polarization camera for computer vision with a beam splitter,” partment, JHU, while he was also a member of the J. Opt. Soc. Amer. A, vol. 11, no. 11, pp. , 2935–2944, 1994. professional staff at the Johns Hopkins Applied [50] M. Schadt, “Liquid crystals in information technology,” Phys. Chem., Physics Laboratory, JHU. He became an Assistant vol. 97, no. 10, pp. 1213–1236, 1993. Professor of electrical and computer engineering in [51] H. Povel, “Imaging stokes polarimetry with piezoelastic modulators and 1989, Associate Professor in 1993, and Professor in 1996 at JHU. In 1995 charge-coupled-device image sensors,” Opt. Eng., vol. 34, no. 7, pp. and 1997, he was a Visiting Associate Professor and Visiting Professor, 1870–1878, 1995. respectively, in the Computation and Neural Systems Program, California [52] K. Thompson, D. Rust, and H. Chen, “A compact polarization imager,” Institute of Technology, Pasadena. In the Summer of 2001, he was a Visiting Johns Hopkins APL Tech. Dig., vol. 16, no. 3, 1995. Professor at Tohoku University, Sendai, Japan. In 1999, he was appointed as [53] Z. Kalayjian and A. Andreou, “Integrated 1-D polarization imagers,” the first Chair and Director of the Computer Engineering Program, JHU. He in Proc. 29th Annu. Conf. Information Sci. Syst., Baltimore, MD, Mar. is the cofounder of the Center for Language and Speech Processing, JHU. He 1995, pp. 163–168. is a coeditor of Adaptive Resonance Theory Microchips (New York: Kluwer, [54] K. Miyasaka, “Iodine complexes: formation, structure, and properties,” 1998). He is associate editor of the Journal Neural Networks. His research Adv. Polymer Sci., vol. 108, 1993. interests include integrated circuits, sensory information processing, and neural [55] H. Takamiya and Y.Tanahashi, “On the poly (vinyl alcohol)-iodine com- computation. plexes,” J. Appl. Polym. Sci., vol. 50, pp. 1807–1813, 1993. Dr. Andreou is a recipient of the National Science Foundation Research Initi- [56] M. Seto and Y. Meada, “Light polarization in iodine-doped polyvinal ation Award and the 2000 IEEE Circuits and Systems Society Darlington Award. alcohol films,” Hyperfine Interactions, vol. 68, pp. 221–224, 1991. [57] J. Guo and D. Brady, “Fabrication of high-resolution micropolarizer array,” Opt. Eng., vol. 36, no. 8, pp. 2268–2271, 1997. Zaven Kevork Kalayjian was born on December 26, 1969, in Union City, NJ. He received the B.E.E. from the University of Pennsylvania, Philadelphia, in 1992, and the M.E.E. and Ph.D. degrees from The Johns Hopkins University, Baltimore, MD, in 1994 and 1999, respectively. At present, he is with Epigy Labs, Yerevan, Armenia. He has done research on a variety of topics related to analog and optoelectronic VLSI, including silicon retinas, floating gate circuits, microelectromechanical systems, and VCSEL/sil- icon hybrid architectures.