Figure 1, 4 Transistor DRAM
Figure 2, Multiple Quatum Well
Figure 3, Bandgap Diagram
Figure 4, Absorption Diagram for the SEED
Figure 5, SEED
Figure 6, S-SEED
Figure 7, VCSEL
Figure 8, Optoelectronic XOR
Figure 9, FCSEL
Figure 10, LAOS device
Figure 11, Bitonic Sorter Diagram

A general overview of smart pixel technology is presented. A discussion of the definition of "smart pixel" is presented. The major concerns effecting the progression of smart pixel technologies are examined. Applications for and architectures of smart pixels are discussed.
 
 
 
 

    In starting the research for this paper I was overwhelmed. The wide variety of technologies that have been utilized in smart pixels makes the general overview smart pixels almost impossible. It was difficult to find a unifying theme in any of the papers I researched.
    The name smart pixel is combination of two ideas, "pixel" is an image processing term denoting a small part, or quantized fragment of an image, the word "smart" is coined from standard electronics and reflects the presence of logic circuits. Together they describe a myriad of devices. These smart pixels can be almost entirely optical in nature, perhaps using the non-linear optical properties of a material to manipulate optical data, or they can be mainly electronic, for instance a photoreceiver coupled with some electronic switching. This leads to the question; How smart is smart? How much logic must a device contain to be considered smart, and for that matter, how much of the device must be optical in nature to be considered a pixel?
    The potential uses for smart pixels are almost as varied as are the designs. They can be used for image processing, data processing, communications, and that special sub-niche of communications, computer networking. While no immediate commercial use for smart pixels has risen to the forefront, smart pixels systems are utilizing technology developed for a wide variety of other commercial applications. As lasers, video displays, optoelectronics and other related technologies continue to progress, it is inevitable that smart pixels will continue to ride on the backs of these commercially successful technologies. At the same time smart pixel or smart pixel-like systems have become available on a limited basis for a wide range of applications such as adaptive optics.
    At first smart pixels were envisioned as a better way to carry out image processing, an alternative to standard digital image processing which involves converting an image to digital data, manipulating the data, then converting back to an image to be displayed. The use of smart pixels to process an optical image may eliminate or simplify some of those steps.
    Optical data processing can be seen as the exact opposite, or mirror image if you will, of digital image processing. Instead of converting an image to digital data; Optical data processing involves converting digital data into an image, using optics to manipulate the image, then converting that image back to digital data. In this example I am not using the word image in the sense that it will be viewed by a human, but image in this case means a two dimensional array of optical data.
    It is common for researchers who are unfamiliar with smart pixels to ask "Why?". As before, the answers can be as varied as the designs being considered, and the problems being attacked. In the area of data processing smart pixels lend themselves well to vast parallelism. While the band width of any one pixel may be the same as, or even less than that of a standard electrical connection and electronic logic, that single smart pixel could be part of an NxN array of four thousand smart pixels, leading to a redefinition of bandwidth. (2)
    The actual physical structure of these devices can be overwhelmingly varied, but two main choices stand out. Modulators vs. Light Sources, in some of the smart pixel designs the light is modulated by be means of a modulator, such as a Multiple Quantum Well (MQW). Other designs involve light modulation by means of switching the light source on and off. In research currently underway at the University of Cincinnati, micro mirrors are being used to modulate the light, leading to an Opto-Electro/MEMS smart pixel system.
    As with all engineering endeavors the design of a smart pixel will suggest a prudent application, and conversely the presence of an application will lead to the development of a smart pixel system that is well suited to the problem at hand. Almost always the smart pixel system can be viewed as an interconnect system. The speed of optical signals in free space leads to an independence of interconnect path length and a freedom from the cumbersome RCL time lags of conventional electronics, while the fact that optical signals do not scatter off of each other allows interconnect pathways to cross. Useful three dimensional interconnect systems have been constructed in which smart pixels and conventional optics work in tandem. All smart pixel systems pose a unique optical processing problem, and so smart pixels remain inseparable from their optics. 
    This reasoning led to an obvious use of complex interconnection architectures. The use of the word complex does not merely connotate a complicated pattern, but in this case it means that the interconnect pathways serve to implement a specific algorithm, such as the Perfect Shuffle (1). While complex interconnects utilize some of the inherent strengths of optical systems, they are also very prone to loss of signal, and as I will discuss later, loss of optical power can lead to a reduction of system speed.
    A smart pixel is an optoelectronic device that may posses logic, memory, electrical connections between neighbors or global connections, on chip sensors or transducers, and/or optical output. This makes smart pixels incredibly varied in form and function.
 
 

    Spatial light modulation for smart pixels can be implemented by electrooptical, acoustooptical, mechanical, photorefractive, electroabsorbtive, and optical absorptive means. This list is probably incomplete.
    Of the many modulation devices available two deserve mention and must not be omitted from any comprehensive study of smart pixels. They are the Multiple Quantum Well and Liquid Crystal.

    The study of liquid crystals is a science unto itself and a complete examination of LC technology, even the topics related to smart pixels, would be beyond the scope of this paper. The LC systems used in smart pixel applications resemble the LC displays used for television and computer screens.
    Liquid crystals are a material that can attain reflective or absorptive states, under the influence of an electric field. This is a very simplistic description, in reality, the LCs can attain more than two states, and can have complex switching characteristics, but this simple description will suffice for now. A liquid crystal display for use in smart pixel applications can be viewed as a two dimensional array of switchable mirrors. These mirrors either reflect, or do not, under the control of some electronic logic.
    A smart spatial light modulator (SLM) can be constructed by using LCs in conjunction with an array of memory circuits. The memory can be addressed in a standard fashion and data can be written onto the memory, then the data is displayed on a LC SLM which is positioned directly over the memory (3).
    As stated earlier the LC display forms a two dimensional array of switchable mirrors. Each mirror in the array can be addressed and controlled in one of three methods;
    1. Direct addressing. The direct addressing is the simplest. Each mirror is addressed with a direct metal connection from the corresponding memory. This requires a large number of electrical connections and results in a size limitation for the array.
    2. Passive Addressing. In passive addressing the mirror array is divided into rows and columns. One column is enabled at a time and data is written to the individual rows. Each column is enabled in succession until the entire display has been updated.
    3. Active Addressing. In active addressing some switching logic is present on board the LC display. This speeds up the addressing process, makes more current available to switch the mirror, stabilizes the state of the mirror, and allows the entire array to be updated within the switching time of a single mirror element.
    These addressing schemes are exactly the same as one would encounter in video display applications. This analogy to video displays is useful in illustrating the strengths and weaknesses of LC smart pixels and smart SLMs. Anyone who has used a lap top computer, and been confused by the mouse cursor leaving tracers across the screen, realizes that LCs can have very sluggish switching times. In the smart pixels and smart SLMs the pixel size can be very small. Since switching time is proportional to size, this size reduction serves to speed up the switching time drastically.
    One can view an element of a LC display as a capacitor with complex dielectric, the liquid crystal material, between the plates. This dielectric switches states when an electric field is applied. The stronger the field , the faster switching can occur. Reducing the size of the mirror element is like reducing the distance between the capacitor plates. This results in a greater Volts/cm, or increased E-field. If the voltage across the capacitor is increased the switching time will decrease.
    So while LCD smart pixels and smart SLMs may have poor switching time now, the unstoppable march toward miniaturization, and the move toward a greater range of available voltage in micro electronics may serve to increase the switching time of LCs drastically. LCs have excellent optical contrast between the on and off states, which can make them desirable over other, faster technologies. It is certain that the maturity of the Liquid Crystal industry will lead to continued research into this area and may lead to commercially viable systems in the future. 

Figure 1. 4 transistor DRAM controlling a LC mirror shown here as CLC. The pixel is driven by means of an inverting buffer, allowing unlimited charge to drive the LC. From Smart Spatial Light Modulators Using Liquid Crystal on Silicon, K. Johnson and D.J. Mcknight, IEEE, 1993 
 
 
 
 

    Multiple Quantum Wells are structures which are fabricated from III-V compounds. They consist of thin layers of material with varied bandgaps and doping levels. Since the MQW is be grown in the intrinsic region of a PIN diode, the structure can serve as both photo detector and light modulator. One side of the MQW is coated with a metal that will form a reflective surface. These MQWs are then commonly flip chip mounted to silicon based logic (2). See figure 2.

Figure 2. A MQW shown with the reflective metal layer in blue, and an optical signal reflecting. If the MQW were in the absorptive state, no reflection would occur.
 
 

    MQW structures used in smart pixel applications exhibit the Quantum Confined Stark Effect. A full understanding of this effect is well beyond the scope of this paper and therefore will not be discussed at length. It will suffice to say that the Quantum Confined Stark effect causes the MQW to have optical absorptive properties that vary with the strength of an applied electric field (4).
 
 

    In the SEED the properties of MQWs can be used to directly implement logic, creating and optical logic gate. When a bias is applied to the MQW the band diagram tilts, as shown in figure three.

Figure 3. A simplified bandgap diagram for a MQW showing the tilting when bias voltage is applied.

This tilting of the band structure causes the absorption properties of the device to change. A plot of absorption vs. wavelength is given in figure 4. 

Figure 4. The absorption curves of a SEED device for three different bias voltages showing the radically different absorption properties for light of wavelength l_o  at these voltages.
 
 

    Since the MQW structures reside within the intrinsic region of a PIN diode, absorption can cause electron hole pairs to be produced, resulting in a diode current. If the device is biased and put in series with a resistor as shown in figure 5, you obtain a non-linear device. The current produced by the SEED creates a voltage to appear across the resistor, which effects the bias of the SEED causing its absorption properties to change.

Figure 5. A SEED in series with a resistor. Optical input causes the generation of a current which in turn effects the voltage across a resistor, this effects the bias of the SEED and thereby effects the absorption.
 
 

    This effect can utilized to form optical logic gates. One SEED may be used to drive another SEED as shown in Figure 6. Optical input , r and s are directed into the two SEEDs, one SEED drives the other and the output is dependent on the relative intensity of the two signals. This is known as the Symmetric SEED or S-SEED configuration. This can be used to implement such logical operations as the r-s flip flop. Other operations are possible such as NOR, OR, NAND, and AND, along with various memory functions.

Figure 6. Two SEEDs shown in the S-SEED configuration implementing a rs flip flop. From An Introduction to Photonic Switching Fabrics, H.S. Hinton, Plenum Press, 1993. 
     SLM based smart pixels can take on many more forms than those presented here, but MQWs and LCDs are excellent representatives of a large class of smart pixels.

    In a discussion of light source modulated smart pixels, it is necessary to understand the devices that produce the light. The Vertical Cavity Surface Emitting Laser (VCSEL) is a very important and useful light source. Other recent developments have involved the use of Folded Cavity Surface Emitting Lasers (FCLELS) and non lasing sources such as diodes.

    VCSELS utilize a quantum well structure to confine charges to an active region much like edge emitting lasers. The main difference between VCSELS and other semiconductor lasers is the vertical structure. Most semiconductor lasers are planar and emit out of the edge facet on all sides. This configuration allows more active region than in VCSELS. The vertical lasers are constructed from the same planar epitaxy method as the edge emitting lasers, then etch back is used to produce a cylindrical structure. Because the light spends a relatively small amount of time in the gain region, it is necessary to optimize the cavity. Layers are grown such that they form Bragg planes so that light with the desired wavelength is preferentially propagated. This structure is illustrated in figure 7.

Figure 7. A very simple depiction of a VCSEL showing the substrate, layers of GaAs and AlAs that form the Bragg planes, the quantum well region where gain occurs, the p and n doped regions that make the p.n. diode junction.
 
 

    The VCSEL is crucial to smart pixel applications because of the ability of VCSELS to form two dimensional arrays. They are constructed out of material that is convenient for fabrication of photodetectors and in some cases logic, so, devices like VCSELS and FCSELS can be utilized in monolithic smart pixels.
    Smart pixels that utilize lasers have an excellent contrast between the on, and off states. This contrast comes at a significant cost in power consumption. This power consumption can cause problems with heat and may reduce switching speed. VCSEL based, and for that mater any surface emitting laser based smart pixel can be used to implement just about any logic function imaginable. In Figure 8 a VCSEL is used to implement an XOR gate and in Figure 9.

Figure 8. An XOR implemented with 1 VCSEL and 4 phototransistors. Taken from Optoelectronic XOR Using Hybrid Integration of Phototransistors and VCSELS, F.R. Beyette, IEEE 1993. 
 
 
 

Figure 9. A monolithic FCSEL smart pixel. Reprinted from High Speed Optoelectronic InP/InGaAs Logic Pixel, Dong-Su Kim, C.Dries, Milind Gokhale, International Conference on Indium Phosphide and Related Materials, 1997.
 
    Other devices have been proposed as substitutes or lasers. In one such instance a system utilizing a heterojunction phototransistor (HPT) in vertical integration with a light emitting diode comprising a structure named the Light Amplifying Optical Switch (LAOS). The LAOS is a bistable device that at low applied voltage, behaves like an HPT diode pair, but at high voltage, switches to a large current by means of a region of deferential negative resistance. Figure 10 shows the structure of a LAOS device and a smart pixel logic gate that can be constructed with such a device. 

Figure 10. A LAOS device structure with logic circuit gate for smart pixel.

    The LAOS shows good on-off contrast, and is capable of operation in a broad band of wavelengths, freeing the designer from the need for a stable wavelength light source. Unfortunately, the switching time for these devices is relatively poor when compared to other smart pixel systems.

    Up to this point I have covered smart pixels with an emphasis on the individual pixel unit. The power of smart pixel technology is its ability to harness the advantages of optical interconnects, whether to simply cultivate massive parallelism, or to implement complex interconnect architectures (5), the versatility of the smart pixel becomes evident when viewed on a system level.
 
 

    The perfect shuffle is a way of manipulating data that lends itself well to certain tasks like sorting. In theory, the perfect shuffle is the fastest way to sort data. The concept of a perfect shuffle is intuitive to anyone who has ever played cards. Imagine a deck of cards, cut the deck exactly in the center, then shuffle them such that the two halves of the deck interleave perfectly. There is a specific mathematical representation for this that is not important for this paper. The deck of cards is symbolic of a linear array of numbers or words. The concept of a perfect shuffle can be extrapolated onto a two dimensional array, resulting in what is called a folded perfect shuffle.
    The perfect shuffle, when coupled with a specific bitonic sorting unit can form the most efficient means of sorting data. The concept of a bubble sort is familiar to anyone who has a rudimentary understanding of computer programming. If you have a list of words and you wish to sort those words, the bubble sort uses at least X2 clock cycles, where X is the number of words. By using the perfect shuffle in conjunction with a bitonic sorter, this same task can be accomplished in no less than Log(X) clock cycles.
    This sorting network was first visualized in the nineteen seventies by a man named Batcher, but it was realized at the time that conventional computing techniques would not be able to accomplish the task. Figure 11 depicts a Bitonic Sorter. This is a black box representation of a device that compares two values and exchanges them. It outputs the two input values in one of three ways, it can output the greatest value to the top output, and the lesser value to the lower output, this would be a sort up. or the reverse of that, sort down. It can also pass the values strait through without any compare and exchange taking place.
    As shown in the diagram, word one has a long electrical connection, word two has a short electrical connection. In an actual sorting network, many of these individual units would be chained together, functioning in parallel. At any given clock cycle, the execution of the perfect shuffle could result in electrical connections like the ones shown in the diagram. The difference in time lag between the two connections can slow the computer down so much that the advantage of speed is lost.

Figure 11. A bitonic sorter. Word 1 has a long electrical path to reach the sorting unit, Word 2 has a short electrical path. Also shown are two control inputs necessary for operation. This sorting unit will output the word with greater value to output one and the lesser word to output two if it is in the "sort up mode". It can also sort down or pass the data without performing a compare and exchange.
 
     Parallel systems like this that take advantage of complex interconnects were not feasible until optical interconnects arrived. The implementation of an optical perfect shuffle bitonic data sort has been accomplished many times and has become the archetypal test case for optical computing systems.
    The problem with complex interconnect systems is that they tend to lose light power. Every time a beam of light hits a mirror, passes through a lens, gets passed into a fiber, or manipulated in just about any way, there is loss of signal. If this loss requires a higher power light source, switching speed may be diminished. 
    Simple interconnect systems can be vary useful and easier to construct and align. A good example of a simple interconnect system is a board to board optical interconnect using either light emitters, or SLMs.

A comprehensive discussion of smart pixels leads one to consider computer architecture, communications, and device electronics to name a few. The innovators of smart pixel technology will continue to utilize ever imaginable breakthrough until finally commercially viable systems are realized. I envision a near future where such diverse technologies as Micro Electro Mechanical Systems (MEMS) and Non Linear optical materials contribute to this development.
 
 

(1) H.S. Stone, "Parallel Processing with the Perfect Shuffle" IEEE Trans. in Comput. Vol.2 1978

(2) A. Lentine, K. Goosen, "High Speed Optoelectronic VLSI Switching Chip with > 4000 optical I/O Based on Flip Chip Bonding of MQW Modulators and Detectors to Silicon CMOS", IEEE Journal of Selected topics in Quantum Electronics Vol. 2, 1996

(3) K.M. Johnson "Smart Spatial Light Modulators Using Liquid Crystal On Silicon" IEEE Journal of Quantum Electronics, Vol. 29, Feb 1993.

(4) H. Hinton "Progress in smart pixel Technologies" IEEE journal of Selected topics in Quantum Electronics, Vol2. April 1996.

(5) F.A. Tooley, "Challenges in Interconnections Electronics" IEEE Journal of Selected topics in Quantum Electronics, Vol.2 April 1996.

(6) F. Beyette, K. Gieb "Optoelectronic XOR Using Hybrid Integration of Phototransistors and Vertical Cavity Surface Emitting Lasers" IEEE Photonics technology Letters, Vol. 5, Nov. 1993.

(7) C. Williamson, F. Beyette, "Smart Pixels Using Light Amplifying Optical Switch (LAOS)" IEEE Journal of Quantum Electronics. Vol. 2 Feb. 1993.

(8) J. Cheng, ping Zhou, Surface Emitting Laser-Based Smart Pixels For two-dimensional Optical Logic and reconfigurable Optical Interconnects", IEEE Journal of Quantum Electronics, Vol. 29, Fe. 1993

(9) H. Hinotn, " An introduction o Photonic Switching Fabrics" Plenum Press, New York, 1993.

(10) J.S. Kane "Realizing optical logic with a smart pixel spatial light modulator." Applied Optics, Vol. 35, No. 8 10 March 1996

(11) D.M. Chiarulli, "Optoelectronic Cache memory system architecture" Applied Optics, Vol. 35, No 14, May 1996.

(12) M.P.Y. Desmulliez, "Performance analysis of self electro optic effect device based (SEED-based) smart pixel arrays used in data sorting. Applied Optics, Vol. 35, No. 32 Nov. 1996