Nonvolatile storage in photorefractive crystals.

Holographic storage in photorefractive crystals'` 3 has a high storage density owing to three-dimensional storage and a high readout rate owing to parallel retrieval. The practical development of such memories has been impeded by a lack of materials that have all the suitable properties. Specifically, the most difficult requirement to satisfy is the simultaneous need for materials that are maximally sensitive to light in order to increase the recording speed and the need for a nonvolatile memory that is not affected by illumination to light during readout. Thermal' or electrical 4 5 fixing are among solutions that were demonstrated to address this problem. Thermal fixing is the most commonly used method now, but it is difficult to design practical systems with this approach. Moreover, it is incompatible with a reprogrammable memory. Electrical fixing is in principle more compatible with a practical reprogrammable memory; however, it is still not well developed. Another approach is the use of periodic copying. 6-9 With this approach the contents of the hologram are refreshed, preventing decay owing to readout. We can also use materials such as strontium barium niobate, whose absorption varies significantly as the polarization is rotated.10 Holograms can then be recorded with one polarization and read out with an orthogonally polarized beam that is less absorbing. This introduces a write/erase asymmetry that can delay significantly the erasure of the holograms. In this Letter we describe a nonvolatile holographic memory that employs different wavelengths in the recording and readout phases. A dual-wavelength scheme was used previously by McRuer et al." to implement nonvolatile optical interconnections with a photorefractive crystal. Pauliat et al.' used the dual-wavelength method to construct a deflector. Kfilich13 devised a dual-wavelength storage scheme in which spherical waves are used to construct holograms. In his scheme he could read out a single hologram at a different wavelength by changing the sphericity of the reconstructing wave. In the dualwavelength memory that we describe, plane-wave references are used instead, and multiple holograms can be stored and recalled. The recording wavelength Al is selected near the peak of the trapinduced absorption spectrum of the material. The readout wavelength A2 is chosen in a region where absorption is as small as possible. Typically the recording wavelength is shorter than the readout wavelength. The large drop in absorption away from the absorption band yields a very large write/erase asymmetry. For example, we recorded holograms at Al = 488 nm and exposed the hologram to A2 = 632.8 nm light at a non-Bragg-matched angle for approximately 20 h at intensities of 0.25 mW/cm2 without appreciable change in diffraction efficiency. If a hologram consisting of a single grating is recorded, then the hologram can be Bragg matched at any wavelength if A2 < A/2, where A is the grating fringe spacing. When the hologram consists of multiple gratings, however, readout becomes more complex, because each grating requires a distinct Bragg-matching condition. We describe a method for circumventing this problem so the information recorded at Al can be read out at A2. The recording geometry is shown in Fig. 1. The information to be stored is presented from a twodimensional spatial light modulator (SLM; a liquidcrystal television in the experiment). Lens Li takes the Fourier transform of the input image, and a hologram is formed by the introduction of a plane-wave reference beam. Consider first the case in which a single hologram is recorded. The hologram is equivalent to a superposition of sinusoidal gratings, with each grating corresponding to a pixel on the SLM. We would like the readout beam to be simultaneously Bragg matched to all the gratings so that the entire recorded hologram can be efficiently reconstructed by the A2 beam.Unfortunately this is not possible. We explain this with the aid of the k-space diagram shown in Fig. 2. The recording reference beam is drawn as a single vector, RI, whereas the signal beam coming from the SLM is drawn as a cone of k vectors. A representative wave vector, Si, is shown in Fig. 2. The hologram itself consists of grating vectors connecting the tip of the reference k vector to all the points of the signal cone k vectors on the A, k sphere. The grating vectors are not drawn in Fig. 2. To consider what happens when we illuminate the hologram with a A2 plane wave, we need to draw a second A2 k sphere, which we assumed to be smaller in Fig. 2. This plane wave is represented by the R% wave vector in Fig. 2. For clarity we have assumed that the hologram is illuminated by the A2 beam from the opposite side, but the hologram can be read out from either side. The intersection of the