Smart Time-Domain Optical Memory

For the 15 years SRI has been investigating the physics and technology needed to develop high-performance coherent time-domain optical memory (CTDOM) and optical processing systems based on the concept called the stimulated echo. Very high storage densities are possible, 1000 gigabits per cubic centimeter or more. Single-channel data rates could be more than 10 gigabits per second. Even faster data throughput could be accomplished by massively-parallel storage, retrieval, and in-memory processing of holographic images.

Under a contract from the Defense Advanced Research Projects Agency (DARPA) through the Air Force Office of Scientific Research (AFOSR) we investigated critical issues that need to be addressed before the CTDOM technology can be used in practical applications:

Development of high density terabit rapid-access random-access optical memory
System architecture and algorithms that make efficient use of the time-domain memory technology for use as smart memory
Materials issues related to system performance

Physical Principles

The recording medium consists of an inert crystal host (such as Y₂SiO₅), with a low concentration of rare-earth impurity ions (such as Eu³⁺), distributed randomly within the crystal. Impurity ions at the various substitution sites have slightly different crystal-field environments, which results in small spectral shifts that are distinguishable at cryogenic temperatures.

Information is stored in the impurity ions by a mechanism called hole burning. Relaxation of optically excited ions results in a nonstatistical redistribution of population among the nuclear-spin or hyperfine sublevels. Ions that absorbed light previously have a different (usually smaller) probability of subsequent absorption. Memory of the history of optical excitation is eventually lost through nuclear-spin relaxation, which may take as long as 24 hours, depending on the nature of the crystal host material and the storage temperature.

When the crystal is illuminated by a complex time-sequence of laser pulses, each impurity ion sees the pulses as either in-phase (constructive interference, leading to more absorption) or out-of-phase (less absorption) with each other, depending on the ion's characteristic absorption frequency, the time separations, phase differences, and spectral composition of the laser pulses. Ions at different spatial locations in the crystal record different information because of time-of-flight evolution of optical phases and because the laser pulses may have different directions of propagation.

The recorded information is recalled by a mechanism called the stimulated echo. When the crystal is subsequently illuminated by another laser beam, called the "read" pulse, the spatial and absorption-frequency modulation of nuclear-spin population produces directional and temporal interferences between the emissions of the impurity ions. The result is spatially collimated bursts of laser-like coherent emission that constitute an "echo" of the previous history of optical excitation.

We model this mathematically by the equation

E₄(x,y,t) = E₁(x,y,t) # E₂(x,y,t) * E₃(x,y,t)

Where the symbols "#" and "*" indicate Fourier correlation and convolution, respectively. The first two laser pulse streams, represented by their electric fields, E₁(x,y,t) and E₂(x,y,t), and called the "write" and "data" pulses (or "data" and "write," depending on which pulse stream contains the variable information to be stored). The third laser pulse stream, called the "read" pulse, is represented by E₃(x,y,t). The resulting "echo" output pulse stream is represented by E₄(x,y,t).

Experimental Description

Initial experiments used high power pulsed dye lasers. The write and read beams consisted of single short pulses. The data beam was a sequence of time-position-encoded bits or pulses generated by manually adjusted beam splitters. Two-dimensional images were generated by transmission through static masks (typically 35 mm slides). Current experiments use moderate-power continuous-wave (CW) dye lasers. Any or all of the three input beams can be chopped, attenuated, or phase modulated with acousto-optic (AOM) or electro-optic (EOM) modulators. The dye laser frequency can be swept or randomly positioned in any frequency bin inside the inhomogeneous bandwidth of the impurity ions. Two dimensional images are generated by transmission through a spatial-light-modulator (SLM). The modulators and laser tuning are all computer controlled. Recalled images are recorded by a CCD camera and captured by a fast frame grabber.

Optical Image Recording, Processing, and Pattern Recognition

By encoding information on the spatial profile of the laser beam we are able to store multiple two-dimensional images. If the reading laser pulse has a uniform spatial intensity distribution, then faithful copies of the stored images are retrieved. If the reading laser pulse also carries information in the form of a nonuniform spatial intensity distribution, then the retrieved or echo images are modified or "processed." The lens that focuses the laser beams into the crystal performs a spatial Fourier transformation on the intensity profiles. An inverse Fourier transformation is performed on the echo images, resulting in a correlation or convolution of the information contained in the original data and read image pulses.

We have recently demonstrated the storage and retrieval of 500 wavelength-multiplexed holographic images, each consisting of 512x488 pixels, at the video frame rate (30 Hz). The rate of image storage is currently limited by computer hardware and the spatial light modulator (SLM). Even with the low laser peak power transmitted by the SLM, only about 7 mW, the time needed to store a single image was less than 75 microseconds. This suggests that recording/retrieval rates in excess of 14,000 frames per second are achievable.

Serial Bit-Stream Storage and Processing

If the first (E₁) and third ("read" or E₃) laser pulse streams are identical, the correlation E₁ # E₃ results in a delta function, and the "echo" pulse stream (E₄) is a faithful copy of E₂. On the other hand, if E₃ = E₂, then E₄ is a time-reversed copy of E₁. By using a complicated amplitude- and phase-modulated "read" pulse stream we can perform a variety of in-memory processing functions and can improve memory performance.

For example, multi-bit biphase-modulated pseudorandom codes or Barker codes can be used in both the "write" and "read" beams to improve fidelity by avoiding spectral congestion and reducing the effect of laser frequency drift. If the "write" pulse stream is modulated both in amplitude and phase, the stored data is effectively encrypted. Data thus stored can be retrieved only if the exact write pulse modulation code is known. One important feature associated with this approach is that the length of the code can be arbitrary, permitting the storage of ultra-wide bandwidth data with the modest laser power available with commercial lasers.

Another example is analysis and recognition of optical data streams by spectral holography. In this scheme, the incoming signal is rapidly analyzed by performing simultaneous correlation operations with many reference signals stored in advance through angular multiplexing. Recognition is accomplished by identifying the direction that yields the maximum correlation signal.

Principal Investigator

David L. Huestis (E-mail) (CV) (Publications)

Representative Publications and Patents

M.K. Kim and R. Kachru, J. Opt. Soc. Am. B 4, 305 (1987).
M.K. Kim and R. Kachru, Opt. Lett. 12, 593 (1987).
E.Y. Xu, S. Kröll, D.L. Huestis, R. Kachru, and M.K. Kim, Opt. Lett. 15, 562 (1990).
X.A. Shen and R. Kachru, Opt. Lett. 17, 520 (1992).
R. Kachru, E. Y. Xu, S. Kröll, D. L. Huestis, and M. K. Kim, U.S. Patent 5,204,770 (April 20, 1993).
R. Kachru, Y.S. Bai, and Y.A. Shen, Adv. Mater. 6, 791 (1994).
X.A. Shen, E. Chiang, and R. Kachru, Opt. Lett. 19, 1246 (1994).
Y.S. Bai and R. Kachru, U.S. Patent 5,369,665 (November 24, 1994).
X.A. Shen, Y.S. Bai, E.M. Pearson, and R. Kachru, U.S. Patent 5,381,362 (January 11, 1995).
X. A. Shen, A.-D. Nguyen, J. W. Perry, D. L. Huestis, and R. Kachru, Science 278, 96 (1997).
A. D. Nguyen, X. A. Shen, D. L. Huestis, and R. Kachru, Algorithm for Extraction of Page-Formatted Binary Digital Data," Applied Optics 37, 8215-8218 (1998).

Related Web Pages

Send comments and suggestions to david.huestis@sri.com

URL: http://www-mpl.sri.com/project/pyu5310.html