|
 |
Holographic storage: are we there yet?
Glenn T. Sincerbox
Professor of Optical Science, Director of the Optical Data Storage Center
University of Arizona
Tucson, AZ 85721
phone: 520.621.4260
fax: 520.621.4358
e-mail: sinbox@u.arizona.edu
Introduction
Storage requirements are exploding; our needs appear to be insatiable. We are all pack rats when it comes to information; rarely do we discard or erase anything for fear we may need it someday. Computer storage is the electronic analog of the attic - continually being stuffed with junk - when the attic is full, we move to the basement or rent a storage shed! Even if we were more diligent in periodically cleaning up our disk drives it would probably be a loosing battle; new software tools with greater power and versatility tempt us to dedicate ever increasing amounts of storage and multimedia requirements for text, images, video and audio consume even more. It has been estimated that the requirement for storage in the year 2000 will exceed 12 exabytes (1020 bits). 12 exabytes at 1 cent per megabyte is a 120 billion dollar business. This gets our attention!
This appetite for storage has fueled an enormous growth rate in capacity; magnetic disk storage, for example, has been increasing in aerial density by 60% per year since 1991. IBM recently announced (12/97) that their Research Division has demonstrated a density of 11.6 Gbits/ square inch. New technologies have entered the marketplace; removable storage in the form of Zip and Jaz drives, digital versatile disks (DVD) with promises of 17 Gbytes in four layers on a single disk and near-field optical recording techniques with the potential of 20 Gbytes on a single surface.
These technologies share some common attributes: they are disk-based and hence rely on mechanical rotation and linear storage density to provide data rate; access time also depends on a combination of disk rotation (latency) and linear translation of a head assembly (seek) and finally, the information is recorded as localized, discrete marks on a medium that is thinner than one micron. Faster access, higher data rates and redundant storage of data within the volume of a thicker medium requires a new paradigm in storage. Optics and the basic principles of holography are a particularly attractive way to meet these requirements and may well represent the storage solution of the future.
While many technologies are considered new at the time of commercialization, a large share of them have a history that can be traced back for decades in research and development laboratories. Such is the case for holography. Invented in 1948 by Dennis Gabor as a means to greatly magnify x-ray images1, it was not until the early 1960s with the invention of the laser that the technology became practical for storing and retrieving images and the concepts of holographic data storage were established by Van Heerden2. Armed with this knowledge and with the expenditure of considerable effort by several major corporations, the development of holographic storage went through many cycles only to fall short of becoming a product each time. Despite it’s many promising attributes, holography in general has not created the impact that we expected. We have jewelry, UPC scanners, counterfeit deterrent features, non-destructive testing and pattern recognition systems. While each is an important and valuable application, they have not created major markets for holography. Why, then, is there renewed interest in a 30-year old technology? What has happened to make us use our resources one more time? The answer, of course, is continued progress in the development of the critical enabling technologies, improved material systems and a better understanding of how holographic storage can be used. It has been a long journey and, is often heard on such trips; are we there yet? The response; almost, don’t go back to sleep!
Concepts
In contrast to magnetic and conventional optical recording where an individual bit of information is represented by a highly localized change in some physical property, such as the ablation of a pit or the reversal of a magnetic domain, holographic recording of a single information bit is distributed throughout the entire recording volume. A one-to-one correspondence does not exist between an information bit and a microscopic element in the recording medium. The process by which this occurs is optical interference in which the light from an array (page) of bits, a complex signal beam, interferes with the light in a reference beam to create a three-dimensional pattern of high and low intensity (see sidebar, Physics of holography, showing how a single point is recorded). The recording material samples this 3-D pattern producing a similar variation in some optical property such as absorption or refractive index. This recorded structure contains information about the amplitude and phase information in both beams of light at the time of recording and, upon illumination with a duplicate of only the reference beam, causes light to be diffracted to recreates the signal beam wavefront exactly. The wavefront, representing the page of information, is thus stored and can be retrieved at a later time and can be viewed directly or imaged onto a detector array as if the original object were still in place.
In contrast to conventional holography where light scattered from a three-dimensional scene or object is recorded and can be subsequently reconstructed for visual viewing, holographic storage uses the light coming from a very special binary object known as a spatial light modulator (SLM). As shown in Figure 1, the SLM is used to display a two-dimensional pattern of ‘ones’ and ‘zeros’ which act very much like miniature open and closed shutters representing the information to be stored. (sidebar - a digital information page)
|
|
| |
Figure 1. Two-dimensional page of digital data. A section of a page of binary data, representing what would appear on the spatial light modulator is shown. A typical page would contain 1024 x 1024 bits and the size of each bit cell would be 15-20 microns.
Because the interference pattern is three-dimensional, information about each data bit is distributed throughout the recording volume thereby reducing the sensitivity to material defects that might otherwise obliterate one or more highly localized bits. The effect of a defect is to slightly reduce the signal-to-noise level of all the bits recorded in that volume. While the requirement for a defect-free recording material is less stringent for holography than for the more conventional recording technologies, scattering of light caused by defects and imperfections must be still controlled as this contributes to noise in the reconstructed image.
The effective areal storage density (bits/unit area) can be significantly increased by using a thick recording layer to record multiple, independent pages of data. This should not be construed to mean that each page occupies a different depth in the recording volume, a common misconception, but rather that the holographic structure for one page is intermixed with the holographic structures of each of the other pages. This process is referred to as multiplexing. Retrieval of an individual page with minimum crosstalk from the other pages is a consequence of the volume nature of the recording and its behavior as a highly tuned structure. This effect, known as the Bragg effect, is depicted in the sidebar. The Bragg effect, showing the variation in diffraction efficiency from a holographic structure as a function of a parameter that depends on mismatches in angle or wavelength between recording and playback. The use of angular multiplexing is depicted in Figures 2 and 3 for recording and retrieving multiple pages or holograms. For our purposes it is sufficient to note that for a hologram of high efficiency (i.e., identical wavelength and angle during recording and playback) the efficiency decreases as either the readout angle or wavelength changes and becomes zero depending on the thickness of the recording material. The inset table in the sidebar shows the corresponding angular or wavelength mismatches as a function of recording layer thickness for this point of zero diffraction efficiency. As thickness increases, the recorded structure becomes more highly tuned, much like an interference filter, such that smaller and smaller mismatches can be tolerated. Hence, for example, multiple holograms can be stored in a 1.0 mm thick recording medium in increments of 0.025 degrees. Using the angular technique, as many as 10,000 holograms have been stored in one location of a 6 mm thick lithium niobate crystal3. In a similar 100 micron (1 mm) thick material, the wavelength sensitivity is 9.5 nm (0.95 nm). Wavelength multiplexing is simpler to implement but is highly dependent on the range over which lasers can be tuned whereas angular multiplexing requires a more complex optical system. Today, wavelength tuning can be accomplished over a range of only about 20 nm, too small to multiplex many holograms. Hybrid systems using a combination of angular and wavelength multiplexing have been explored. Several other techniques have also been developed for multiplexing that involve either rotation or translation of the recording medium. These include; rotation of sample about axis normal to plane of incidence, rotation of sample about an axis in plane of incidence (peristrophic multiplexing4) and small translation of the sample over a few microns (shift multiplexing5).
|
|
 |
Figure 2. Fundamentals of holographic storage: recording.
A two-dimensional pattern of ones and zeros is transferred from the computer cpu and is displayed on a spatial light modulator (SLM) as a corresponding pattern of high and low transmission. A laser beam passes through the SLM, picking up the displayed pattern and is directed to a recording medium where it interferes with another beam, called the reference beam, to form the hologram. A new pattern, or page of information, is set up on the SLM and another hologram is recorded in the same volume of material with a different reference beam. As shown here, the reference beam angle is changed with each hologram recording. A large number of different holograms are recorded in this manner, perhaps as many as 5,000, depending on the thickness and properties of the recording medium.
|
|
 |
Figure 3. Fundamental of holographic storage: retrieval.
To retrieve a page of information that had been stored holographically, the recording medium is illuminated with the reference beam corresponding to that page; in this case the direction of the reference beam is selected. This light interacts with all the recorded holograms but is an exact match with only one of them. A beam is light is created that is identical to the light that was coming from the SLM when that page was displayed, illuminated and finally recorded. This light is imaged onto an array of photodetectors, converted to an electrical signal and transferred back to the cpu via the electronic channel. Each page can be retrieved independently by using the associated reference beam.
Another method of multiplexing has been demonstrated that uses the phase of a multitude of reference beams to distinguish one hologram from another6. Referred to as deterministic phase encoding, the individual beams used in angular encoding are incident on the recording medium simultaneously so that each hologram is recorded with the same set of beams. The sets differ from one another by an orthogonal code of phase retardations that have been impressed onto each 'beamlet'. Using a binary code of 0 and 180 degree retardation, unique codes can be constructed that allow the individual holograms to be reconstructed. This method is particularly attractive because access to a hologram can now be achieved by altering the phases of beams and not their direction - a faster and potentially simpler process.
Increased storage density through multiplexing can provide a significant increase in overall storage capacity and is a pathway by which holographic storage will be distinguished from conventional storage.
The 2-dimensional page-oriented nature of holographic storage also utilizes the information capacity of an optical wavefront to allow data to be recorded and retrieved in parallel, a page at a time, rather than serially as in conventional storage. This affords a potential for extremely high data rates subject only to limitation imposed by the I/O devices (SLM, detector arrays) and electronic channels. For example, detector arrays are available today that can run at GHz speeds (using multiple output taps) and it is not difficult to conceive of massively parallel arrays running at effective data rates exceeding 1.0 GByte/sec.
A further distinguishing feature is the speed by which data can be accessed. Relying on movement of light beams and not mechanical mass with devices such as acousto-optic deflectors, or at least very minimal mass in the case of galvanometers, access times in the 10 microsecond to 1 millisecond range can be achieved. True random access can be provided as it is not necessary to pass through other storage locations in accessing a target location. The absence of latency as present in rotating storage systems further enhances access.
Finally, spatial and angular multiplexing provide an opportunity to perform novel functions such as associative retrieval. In this process, the holograms recorded at a given location are illuminated with a partial data page, rather than a particular reference beam, and the complete data page that contains this partial input is retrieved. Analogous to pattern recognition, this promises to provide a very fast search mechanism that locates data by its content rather than its address.
This abbreviated view of holographic storage does little justice to the breadth and complexity of this technology. The intent is not to describe the field in detail but to establish a foundation for the discussion to follow. More information on the subject can be obtained from a recently published collection of key papers7, a book8 and some excellent review papers9,10.
Enabling technologies
Much of what has been described about the technology has been known for over 30 years. What then has happened to cause all the excitement; why are we seeing a new cycle of development and the promise of products within a few years? The answer: holographic storage is dependent on more than just holography; there are many enabling technologies required to make a system and they took time to mature. In contrast to conventional magnetic and optical recording, a holographic storage system is unique because it is not based on one dominant technology but rather requires the integration of many independent technologies. These must be combined in such a way as to provide a system that is both novel and competitive in performance, cost and size. The key components, shown schematically in Figure 4, come from diverse fields including lasers, spatial light modulators, detector arrays, deflection and transform optics, and material science. The list and the figure are far from complete. Optimum performance requires that many of these component subsystems are pushed beyond the current state-of-the-art (e.g., higher speed, more dense SLM's and massively parallel CCD arrays). To do this however, requires more than just the promise of holographic storage for it is not (yet) a sufficient driving force to significantly impact the development plans of companies working on only these support technologies. Thirty years of progress, driven mainly by consumer electronics and the entertainment industry has provided remarkable progress in many of these components. Integrated CCD arrays, developed for camcorders, have replaced discretely mounted photodiodes; liquid crystal spatial light modulators, developed initially for watches and hand-held TVs and subsequently extended to computer output displays, have replaced photographic masks and film transports; and efficient laser diodes, developed for the CD audio market and conventional optical storage, provide sources that can be used directly or as pumping sources to provide blue-green light rather than argon ion lasers. The following three examples demonstrate how rapid the state-of-the-art has progressed.
|
|
 |
Figure 4. Components of a holographic storage system.
Some of the components of a holographic storage system are shown in this simplified schematic diagram.. Not shown are the complexity of optics required, the mechanical transport systems to access other recording volumes, the electronics required to control the system and the channels needed to format and decode data. It will probably occupy the space of a 2-drawer file cabinet.
The light source. The light source for holography must have sufficient spatial and temporal coherence to allow the formation of an interference pattern over the desired volume of space (e.g., throughout the recording medium) and to keep this pattern stationary during the exposure time. With few exceptions, the source is a laser. In general, the laser must have sufficient power at the required wavelength (typically on the order of a few hundred milliwatts cw at the medium); be small enough to fit into a reasonable sized system and, of course, cheap. An argon-ion laser satisfies the first two requirements but falls way short of the size and cost need. A reasonable choice is second harmonic generation of 530 nm light from a diode-pumped Nd:YAG laser. A better choice would be a near-infrared laser diode - these are under development. The ultimate choice will depend on the wavelength sensitivity of the recording material.
The spatial light modulator. The digital information must be organized into a page-like format of ‘ones’ and ‘zeros’ and subsequently modulated onto the object beam as a two-dimensional pattern of brightness and darkness (Figure 1) . The analogy is an array of miniature shutters that are either open or closed. A more practical implementation is a device called a spatial light modulator (SLM). While these devices have been fabricated from many different technologies (electro-optic, magneto-optic, micromechanical, deformographic, etc.) the most common implementation is with liquid crystals. Having their genesis as watches, hand-held televisions and computer displays, considerable engineering has gone into increasing the performance of these devices and driving the cost down through mass markets. A SLM for holographic storage need only be binary and does not require color; it does, however, require high contrast and rapid switching between the on and off states. Contrast levels of 5:1 (on a pixel by pixel basis) appear to be acceptable, and while frame rates as high as 1000 frames/sec will ultimately be required, the recording speed of current holographic materials is such that frame rates of a few 100 frames/sec are adequate. Since mainstream SLM applications only require video frame rates, the current state-of-the-art and low cost cannot be fully leveraged. The pathways to higher speed are there, prototypes are available, additional engineering will be required and it would be nice to have other applications such as optical computing to pull the cost down.
SLMs containing 640 ´ 480 pels are readily available and prototype devices with 1024 ´ 1024 are starting to appear. Typical pel sizes for the former are in the 30-50 micron range and in the 15-20 micron range for the latter.
The detector array. The other end of the optical data channel is a detector array that receives the reconstructed image from the hologram and creates the signals that enter the thresholding, error-correcting and demodulation circuits. Typically a charged-coupled device (CCD), there are several design parameters that directly affect the architecture of the overall system. These include: quantum efficiency of detection at the design wavelength, number and size of the individual detector elements, and data rate. Most commercially available CCDs drop off in quantum efficiency rapidly at longer wavelengths - many have no sensitivity beyond 700 nm - impacting the choice of recording material and laser. Data rates approaching 1.0 Gbit/sec are achieved by segmenting the CCD into sub-areas and reading them out in parallel. As many as 64 channels are in the prototype stage. One of the most difficult design parameters to resolve is the size of an individual detector element. For several reasons, the optimum holographic recording and reconstruction system is a system of unit magnification. The hologram, recorded in the central plane, is the Fourier transform of the SLM and enjoys some degree of translational invariance - important for accessing schemes that require mechanical motion and for removability. In addition, a system with unit magnification (all lens focal lengths are the same) is less sensitive to lens aberrations. These considerations essentially require that the SLM pixel and the detector pixel be the same size, and that 1:1 pixel matching occurs. That is, the SLM is imaged through the recording medium onto the detector such that each SLM pixel and corresponding detector pixel is a separate channel. As can be imagined, alignment is exceptionally critical. Unfortunately, the trend in CCD fabrication is to use smaller and smaller pixels; silicon real estate is expensive, and SLM pixels have a lower limit in size because the cell containing the liquid crystal material can only be made so thin. CCD’s are at 9 microns and getting smaller, SLM’s are limited to about 15 microns. One solution is to demagnify the SLM image to preserve the pixel matching, the other is to oversample the reconstructed image and do some degree of image processing to recover the data. The former has cost and form factor implications, the latter will impact data rate. Both methods are under active investigation.
Control electronics. Figure 4 does not show the enormous complexity of electronic required to drive the optical engine. The serial data stream must have error correction codes (ecc) incorporated, a modulation code added and then formatted into 2-D page-size blocks for writing into the SLM. On the other end, detection of the reconstructed page will generate analog signals the must be thresholded to produce binary bits, the modulation and ecc decoded and, since the data is coming from the detector array in as many as 64 parallel channels, the output data streams must be recombined to produce a meaningful data stream. Detection may also involve considerable image processing to correct magnification, distortion and misalignment of the image onto the detector array. In between, the holographic recording and playback process requires computer control of laser power, multiplexing - angle & spatial, exposure times, calculation of exposure schedules in the case of photorefractive materials, and perhaps real-time servoing of the multiplexing angle and of the image onto the detector array.
Materials:
The requirements. The ideal material for holographic storage should have the properties listed in Table 1.
|
|
 |
The hologram should be formed as local changes in the refractive index of the material so that high diffraction efficiency can be achieved. (see sidebar on recording mechanisms) High optical quality and low scatter are required to insure that the signal-bearing wavefront is not adversely distorted and that the noise level from scattered light is manageable. A thick material is required to use the Bragg effect to its fullest. A large refractive index modulation insures that there is sufficient dynamic range to multiplex the many holograms and the high recording sensitivity allows high speed at reasonable laser powers. From a packaging and reliability viewpoint, the use of diode lasers in the near infrared is preferred. The self processing and fixable requirement go hand-in-hand. If the application calls for a read-only material, then the off-line recording of the holograms permit the use of additional process steps - even wet processing. This, in turn, assures that the holograms are fixed and will not be destroyed upon subsequent reading. The preference, however, is for a read-write material wherein data can be recorded, retrieved and erased as required - similar in performance to magnetic or magneto-optic recording. The requirements, therefore, would be for a material that not only self-develops upon illumination but one that also can be fixed to render it insensitive to subsequent illumination during the recording of additional holograms or the retrieval of data. The fixing process should also be reversible so that the information can be erased and a new hologram recorded. Between these two extremes is an in situ recording process where the information can be recorded but not erased. Referred to as WORM (write-once-read-many), this process has widespread applicability in areas such as medical imagery, satellite telemetry, banking and various legal documents. To meet these requirements, the recording material must have a fixing process that is irreversible - the distinguishing feature between WORM and erasable materials. Long shelf life and inertness imply that the material will remain sensitive over an extended period of time and the hologram, once formed, will not degrade. Finally, the material must be relatively cheap - more on this later.
If holographic storage has had an Achilles heel over the years, it has been the recording material. Certainly, many successful holographic materials have been developed, but these requirements, particularly for self-processing and thickness, greatly reduces the number of choices.
Inorganic recording materials. The most common inorganic materials are ones that exhibit photorefractivity such as lithium niobate (LiNbO3), strontium barium niobate (SBN), and barium titanate (BaTiO3). Lithium niobate has been around for many decades and can be fabricated in large crystals of high optical quality. In the photorefractive effect, the bright regions of the exposing interference pattern excite electrons into the conduction band that drift into the dark regions and become trapped. This redistribution of charge sets up local space charge fields that interact with the host material to alter its index of refraction through the electro-optic effect. The result is a phase hologram that efficiently diffracts light; 100% diffraction efficiency is not uncommon, as these materials can easily be fabricated in thicknesses of a centimeter - which also enhances their ability to multiplex many holograms in the same volume. Obviously, no additional processing is required, but the electronic space charge distribution is easily redistributed by subsequent illumination. This occurs when additional holograms are recorded in the same volume or when an individual hologram is read out. The light from the subsequent exposure of a second hologram redistributes some of the space charge field, partially erasing the first hologram, a third hologram exposure partially erases the first two, and so on. The Nth hologram recorded has partially erased the previously recorded N-1 holograms. This necessitates the establishment of a recording schedule of appropriately decreasing exposure times in order to equalize the final diffraction efficiency of each hologram in the set of exposures. A side effect is to cause the efficiency of an individual hologram to falls off as 1/N2 rather than 1/N. This is fundamental with no means of avoidance except to reduce the number of holograms is a common volume and/or reduce the number of bits on a page. The number of photons per bit must be kept large enough to insure that the bit error rate (ber) does not increase to the point where error-free data recovery is compromised. This is a major limiting factor in the capacity of a holographic storage system.
Unfortunately, this is not the whole story: the illumination necessary for readout of a single hologram also causes partial erasure of all the holograms in that same volume by the same photorefractive effect. Thus the entire assembly of holograms slowly decays in efficiency with each reading until a point is reached when the ber becomes unacceptable. The stored information is highly volatile. The quest therefore has been to develop a fixing process to render the holograms insensitive to subsequent illumination. One such process is to use thermal fixing. The recorded holograms are elevated in temperature to around 120oC and the electronic space charges interact with ions, typically protons, that are more mobile at this higher temperature. The ions move to compensate for the electronic space charge and become locked in place when the temperature is reduced. The ionic charges create their own index variations that are revealed when the readout illumination erases the electronic space charge. This process works quite well, creating a permanent set of holograms with only a small loss in diffraction efficiency. The are, however, some architecture and performance issue with this method as it is a bulk process with long time constants associated with heating and cooling. It is best suited for read-only application where the holograms are recorded at some central facility and then distributed. If reasonable delays between recording and retrieving are allowed, it may also be suitable for use in a write-once storage mode.
Considerable progress has also been made in electrical fixing whereby the electronic space charge is converted to domains in the crystal by applying a field in excess of the coercive field of the material. This is a more rapid process with no thermal time constants involved. As many as 1000 holograms have been fixed by this method11. Work is progressing on reducing the size of the domain structure so that higher spatial frequency holograms can be recorded.
A more elegant solution is to use a gating process during exposure of the hologram; i.e., two stimuli are required to be present for the creation of the hologram and only one is required for readout. The first beam of light (not necessarily laser light) at one wavelength uniformly pumps the material into a long-lived intermediate state (10's of msec) and the second beam forms the hologram at a second wavelength by selectively exciting these electrons into the conduction band where they drift, become trapped and create space charge fields. Readout of the hologram at this second wavelength causes no erasure because electrons can not be raised to the conduction band by this wavelength alone. This process has been demonstrated in LiNbO3 using cw 488nm gating light and 800 nm hologram forming/reading light12. Three-four orders of magnitude in hologram lifetime have been observed. Current efforts are being directed toward reducing the power densities required to produce this effect. Erasure can be accomplished with uniform exposure of both beams. This process has been known for some time but generally required high power pulsed lasers. The new news is that it can now be accomplished with modest cw power. In related work in BaTiO3, dark storage time has been increased to 22 centuries by doping with cerium that produces traps located near the center of the band13.
Engineering solutions have also been devised by which the contents of a hologram are re-recorded immediately after being retrieved. This method can take two forms: (i) the reconstructed image is detected, processed by the computer, feedback to the SLM and re-recorded or (ii) the image is detected and simultaneously used to illuminate an optically accessed SLM and re-recorded. In both cases, the reconstructed bits are thresholded to retain their binary character. The opto-optical feedback of the latter method has been extended to also demonstrate equalization of the efficiency of the recorded holograms, selective erasure of one hologram and recording of a new hologram in its place14. While these solution work, they introduce complexity into the design and impact overall performance, particularly data rate.
Organic recording materials. Holographic recording in organic photopolymer systems has been around for two decades; most of the early interest was directed toward fabrication of holographic optical elements and scanners. Present interest is directed toward optical interconnects and write-once holographic storage15,16. In this material, an organic photosensitive monomer dispersed in a host material starts to undergo polymerization in the high intensity regions of the illuminating interference pattern. As the concentration of monomer decreases locally, concentration gradients are established that cause transport of more monomer into the high intensity areas adding to the polymerized species and leaving behind regions of low monomer concentration in the areas of low intensity. Upon completion of the holographic exposure, the entire area is uniformly illuminated to complete polymerization of remaining monomer and produce a pattern of modulated refractive index. While these are real time materials, they must be fixed and can not be erased. They are particularly suited for WORM applications and are very attractive from an ease of fabrication (crystal growth is a long and complex process) and cost.
Recently, photorefractive effects have been observed in organic systems consisting of nonlinear optical polymers doped with charge-transport agents17. Independent selection of a host material for a large electro-optic effect and a dopant responsible for charge generation, transport and trapping provides greater latitude in molecular engineering and attainment of a high photorefractive figure-of-merit. Since the first demonstration of organic photorefractivity a few years ago, diffraction efficiencies have been increased steadily and now 100% can be achieved.
Organic materials currently suffer from two major drawbacks; as yet, they cannot be fabricated to a thickness greater than 100 microns (hence the number of holograms that can be multiplexed is greatly reduced), and they undergo some degree of shrinkage with exposure, which complicates retrieval of a multiplexed hologram and leads to crosstalk noise. Research into these issues and other polymeric materials continues, however, because of their inherent advantages over grown and polished inorganic crystals.
Optical quality. The importance of the optical quality of the recording material can not be overstressed. It is a significant factor in determining if a data pixel can be registered to a detector pixel over the entire field of the data page. Even with a perfect imaging system, providing exact 1:1 registration, distortions introduced by poor surface flatness, nonparallel surfaces and internal inhomogeneities can rapidly destroy the ability to recover data. An error, in magnification of only a few parts in 104 or an image translation of a few microns has an enormous impact on bit error rate. This source of magnification and translation error could easily be introduced by the material quality.
Data Storage
Photon budget and data reliability. The goal in any storage system is to cram the maximum amount of data into the smallest possible space. If the data can be retrieved with no errors (i.e., the raw error rate before applying error correction and demodulation decoding is zero), the general feeling is that the system design is not aggressive enough. A properly designed demodulation and error correcting code can correct a 10-5 raw bit error rate (BER) to the 10-12 level. Error sources in data recovery will have some combination of Rician (optical) and Gaussian (electrical) statistics, and will require a signal-to-noise ratio (SNR) between 7 and 9 to achieve a 10-5 raw BER.18 Since there are only so many photons to go around, the design must balance the required SNR with the number of bits that a reference beam will interact with at one time. In other words, the number of photons in a reconstructed bit is a function of the number of bits in a page and the number of holograms multiplexed in the same volume. Unfortunately, the diffraction efficiency of a single hologram contained within a set of N multiplexed holograms varies as 1/N2 - hence, there are 106 times fewer photons per bit from a volume containing 1000 holograms compared to a single hologram. Multiplexing 500 hologram pages, each containing 1 Mbit of information (each page coded to contain half ones and half zeros) and illuminating with a 1 watt laser would provide 1000 photoelectrons for detection in 1 msec. This is probably the minimum needed for a 10-5 BER and an instantaneous data rate of 1 Gbit/sec. Variations in other parameters such as quantum efficiency, media figure of merit, fixing efficiency, etc., will alter this number slightly.
Applications
There are no shortages of applications that will require storage of large data files. This is certainly true of the many variants of image files: medical images such as x-rays, MRI and CAT scans; satellite imagery for reconnaissance, surveillance, weather forecasting and geophysical mapping; space exploration; on-line catalogs and manuals; home and commercial still photography and video; and so on. To address these applications, holographic storage should attempt to capitalize on some of the unique attributes of holography. In particular, the storage and retrieval of an entire page of information that may be as large as a megabit in parallel gives rise to data rates that are difficult to achieve with storage devices that operate with serial data streams. Retrieval of a one megabit page in one millisecond provides a data rate of one gigabit per second. These are not unrealistic and have indeed been achieved in the laboratory. If this is coupled with rapid access to multiple pages (images) that are stored in a common volume through one of the multiplexing schemes, these data rates can be sustained over thousands of pages (images). Thus holographic storage is ideally suited for storage and retrieval of large data files such as images. It would be very wasteful and inefficient to use holography to retrieve small files that represent only a portion of a data page.
Integrating holographic technology into a functional storage system is only part of the battle. In the final analysis, any new storage technology must compete in the marketplace with entrenched technologies. Where will the competing technologies be in the year holographic storage is ready - they are truly moving targets! For example, magnetic disc storage is now at $0.07/Megabyte and is expected to be as low as $0.01/Megabyte in the year 2000. If holographic storage is not significantly cheaper, I doubt it will find much more than a niche market as there are very few applications where the customer can be expected to pay a premium for slightly better performance.
I don’t expect to see holographic storage attached to a personal computer; the hardware and media costs do not become competitive with magnetic recording until the storage capacity is over 300 GBytes. This is about $3,000 - much more than the average PC user will pay for storage (unless of course you need 300 GB). It will probably appear as a mini library containing several 100 Gbytes that is ideally suited for image storage, particularly if data rates approaching 1 Gbit/sec can be realized. The first material will be the old standard, LiNbO3, operated in both read-only and write-once modes. Erasable materials will be down the road a bit, as will be their organic cousins.
Summary
The requirements for storage capacity have increased dramatically in the past decade and are expected to continue unabated. At the same time, the storage industry is becoming more diverse; there is no single technology that provides the solutions to all segments of the expanding market. Indeed, there are applications that have specific requirements in capacity, data rate or access speed (for example) that can not be satisfied by the hardware available today. While the limits of performance will continually advance, the requirements will almost always exceed what is available or about to come. In this context, holographic storage holds forth the promise of addressing some of these needs and, as such, can play an important role in a storage hierarchy or as a stand-alone system. The attributes described in this article suggest that parallel channel detector arrays in combination with large data pages can provide very high read data rates and multiplexing schemes with phase-encoded reference beams can provide fast access to large databases.
While progress has been encouraging, at times even exciting, the message contained here is one of cautious optimism. Many of the problems encountered in the early development years have been put to rest. As we dig deeper into systems integration, like the proverbial onion, new layers are uncovered that require innovation and hard work. The central issue is, of course, the identification of a suitable material system. This quest is complicated by the fact that there are different storage applications with different requirements on writeability and erasability, requirements that can not be satisfied by, or forced upon, one single material. Consequently we must sort through the materials, understand their performance tradeoffs and match their expected performance with the target applications. We must then ask if the subsequent storage system design is indeed competitive from a cost performance viewpoint. Does it provide new function, enable new applications and do so at a cost per megabyte that is orders of magnitude better than where the competition will be in the same time frame?
I do not see any fatal flaw in the technology itself; no insurmountable breakthroughs are required. The question will be, pure and simple, can holographic storage compete?
This brief discussion of holographic storage is not intended to be complete but hopefully it is adequate in transmitting some new news about a re-emerging field. The author apologizes to the many colleagues whose work, for lack of space, could not be cited.
References
Figure Captions:
Figure 1. Two-dimensional page of digital data.
A section of a page of binary data, representing what would appear on the spatial light modulator is shown. A typical page would contain 1024 x 1024 bits and the size of each bit cell would be 15-20 microns.
Figure 2. Fundamentals of holographic storage: recording.
A two-dimensional pattern of ones and zeros is transferred from the computer cpu and is displayed on a spatial light modulator (SLM) as a corresponding pattern of high and low transmission. A laser beam passes through the SLM, picking up the displayed pattern and is directed to a recording medium where it interferes with another beam, called the reference beam, to form the hologram. A new pattern, or page of information, is set up on the SLM and another hologram is recorded in the same volume of material with a different reference beam. As shown here, the reference beam angle is changed with each hologram recording. A large number of different holograms are recorded in this manner, perhaps as many as 5,000, depending on the thickness and properties of the recording medium.
Figure 3. Fundamental of holographic storage: retrieval.
To retrieve a page of information that had been stored holographically, the recording medium is illuminated with the reference beam corresponding to that page; in this case the direction of the reference beam is selected. This light interacts with all the recorded holograms but is an exact match with only one of them. A beam is light is created that is identical to the light that was coming from the SLM when that page was displayed, illuminated and finally recorded. This light is imaged onto an array of photodetectors, converted to an electrical signal and transferred back to the cpu via the electronic channel. Each page can be retrieved independently by using the associated reference beam.
Figure 4. Components of a holographic storage system.
Some of the components of a holographic storage system are shown in this simplified schematic diagram.. Not shown are the complexity of optics required, the mechanical transport systems to access other recording volumes, the electronics required to control the system and the channels needed to format and decode data. It will probably occupy the space of a 2-drawer file cabinet.
Side bars:
Physics of holography
When two beams of light are created from the same laser source, and caused to intersect in space(a) , they interact with one another forming something called an interference pattern. In some regions of this pattern, the beams reinforce each others causing an increase in the brightness and in other regions they cancel each other creating areas of little or no light. Depending on the angle between the beams, the oscillations between light and dark can be very close - thousands per millimeter. If one places a photosensitive material into this region of overlap (b), the material becomes exposed where the light is brightest and receives very little exposure in the darker areas. This exposure mimics the interference pattern and is used to create some physical change in the optical properties of the material such as its refractive index, absorption or thickness (c ). Some materials require processing to bring about this change, others do not. The net effect is to create a structure that is identical to the original interference pattern. This is a hologram. If we now illuminate the hologram with one of the original light beams (d), the structure in the hologram will diffract (bend) the light in such a way that it will be identical to the other that was used to form the interference pattern. We have reconstructed the other beam and it can be operated on with lenses and imaged just as if the source of this beam were still there. The figure shows the interference between a parallel beam of light and light from a point source and the subsequent reconstruction of a beam that for all purposes is identical to the light coming from the point source. Any three-dimensional object or scene can be considered to be made up of many point sources - the hologram can be quite complex.
The Bragg effect
The hologram is a very highly tuned structure and becomes more sensitive to any deviations from the recording conditions as it becomes thicker. If during reconstruction, we illuminated the hologram with a beam of light identical to one of the recording beams, then the strength of the diffraction, the efficiency by which we create a new beam, would be a maximum. If, however, the illuminating beam is slightly off in direction or wavelength, the efficiency of reconstruction will be decreased. As the deviation from the recording condition becomes greater, this efficiency will actually fall to zero and rise again to oscillate with ever decreasing amplitude. At the point where the efficiency becomes zero, another hologram can be recorded and the interaction between the holograms will be a minimum. The mismatch in angle ( or wavelength) where this zero in efficiency occurs becomes quite small as the recording medium becomes thicker - as shown in the inset table, it can be as small as a few thousandths of a degree, for example, thus allowing many holograms to be recorded in the same volume.
Mechanisms of hologram formation
Holograms can be formed in very many materials where a diffracting structure, consisting of some change in the optical properties, can be induced with light. Highly efficient holograms can be made in materials that exhibit a change in refractive index, called phase holograms, such as inorganic crystals of lithium niobate and organic photopolymer systems. Holograms of modest efficiency, less than 35%, can be formed as a surface relief or thickness variations in photoresist materials or stamped in plastic like the holograms on a credit card. Finally, holograms can be made in photographic films and plates where an absorption structure is formed from the exposure and development. These are the highest speed recording materials but have efficiencies less than 6%. Many materials form holograms through some combination of the three mechanisms.Table 1. Material requirements.
The ideal holographic storage material should have as many of the following properties as possible.