square surface roughness (R) of Au surfaces and Ag overlayers as a function of scan size after DMG of 250 ML of Ag. After H 2 flame annealing of the Au(111) surface, R over a 5 m by 5 m area was typically in the range of 0.275 to 0.340 nm. After DMG, R was deter- mined for the Ag overlayer (for example, over a 5 m by 5 m area, R = 0.336 nm) and the Au surface over the identical area (for example, R = 0.392 nm) by electrochemically stripping the layer in situ. The larger R of the Au surfaces after stripping of the Ag layer is related to submonolayer roughness that developed from surface alloying [S. G. Corcoran, G. S. Chakarova, K. Sieradzki, J. Electroanal. Chem. 337, 85 (1994)]. 26. The deposition current used corresponds to a flux of 3.1  10 13 atoms cm -2 s -1 (0.02 ML s -1 ). J. D. Porter and T. O. Robinson [ J. Phys. Chem. 97, 6696 (1993)] reported a value of 1.4  10 -6 cm -2 s -1 for the diffusivity of Ag adatoms on Ag(111) in electro- lytes. Calculating the value of (4a 2 D/J) 1/6 , we ob- tained 50 nm as an order of magnitude estimate of the critical island size for our growth conditions. 27. J. Camarero et al., Phys. Rev. Lett. 81, 850 (1998). 28. M. Dietterle, T. M. Will, D. M. Kolb, Surf. Sci. 342, 29 (1995). 29. L. C. Feldman, J. W. Mayer, S. T. Picraux, Materials Analysis by Ion Channeling (Academic Press, New York, 1982), pp. 44 – 46. 30. The yield was determined as the area ratio of the Ag peaks for random alignment and the channeling alignment. 31. The sensitivity of Auger electron spectroscopy and Rutherford backscattering spectrometry is of order 0.5 and 1 atomic %, respectively. See, for example, M. P. Seah and D. Briggs, in Practical Surface Analysis, D. Briggs and M. P. Seah, Eds. (Wiley, New York, ed. 2, 1990), vol. 1, chap. 1. 32. K Sieradzki, S. R. Brankovic, N. Dimitrov, unpublished data. 33. We thank B. Wilkens and R. Culbertson for assistance with the Rutherford backscattering spectrometry measurements and gratefully acknowledge the sup- port of this work by the NSF, Division of Materials Research (contract DMR-9510663). 27 October 1998; accepted 26 February 1999 Strongly Photonic Macroporous Gallium Phosphide Networks Frank J. P. Schuurmans, 1 * Danie ¨l Vanmaekelbergh, 2 Jao van de Lagemaat, 2 † Ad Lagendijk 1 A photo-assisted electrochemical etching technique to fabricate macropores in single-crystalline gallium phosphide (GaP) with variable porosity has been developed. Scanning electron microscopy and x-ray diffraction experiments confirm that the material consists of three-dimensional, interconnected ran- dom networks with pore sizes of about 150 nanometers. Optical transmission measurements demonstrate that the nonabsorbing disordered structures strongly scatter light. The photonic strength is controlled by filling the pores with liquids of different refractive indices. Macroporous gallium phosphide filled with air has the highest scattering efficiency for visible light. In a binary system with components of re- fractive indices n 1 and n 2 , the efficiency of light scattering depends on how these com- ponents are organized in the system, the di- mensions of the components, and the refrac- tive index ratio n 1 /n 2 = m. Scattering of light is strongest if m is large and the length scale of refractive index variation, s, is comparable to the wavelength of light . This regime has received much attention in the search for strongly scattering (photonic) materials. For ordered systems with a periodic variation in the refractive index, that material is a pho- tonic crystal. Such crystals feature photonic band gaps (1, 2): frequency ranges for which light will not propagate in the crystal because of multiple Bragg reflections. If the material is disordered, the interference of scattered light ultimately leads to Anderson localiza- tion (3, 4 ). Photonic band gaps and Anderson localization are closely related. Both inhibit light propagation due to interference (not ab- sorption) and can only be obtained for strong- ly photonic materials, those with m  3 or larger. For infrared light, Anderson localization has been reported for gallium arsenide pow- ders (m  3.5) (4 ), and a two-dimensional (2D) photonic band gap has been reported for macroporous Si (m  3.4) (5). In the visible, considerable progress has been made by the preparation of 3D air-sphere crystals of TiO 2 (m  2.7) (2). Inhibition of the propagation of visible light has not yet been reported for 3D structures—apparently larger values of m (2.7) are needed. Obviously, m is a crucial parameter for photonic materials. In addition to large m, it is thus desirable to have the ability to tune m, allowing investigation of its importance for the photonic strength. We report here on the optical scattering properties of electrochemically etched, macro- porous, single-crystalline gallium phosphide (GaP). The 3D random network of GaP is completely interconnected, as observed by scanning electron microscopy (SEM) and con- firmed by x-ray diffraction. An important con- sequence of the macroporous network is that we are able to modify the refractive index ratio by filling the voids with materials with different refractive indices without disturbing the overall porous structure. GaP has a large refractive index of 3.3 and an indirect band gap of 2.24 eV (550 nm) (6 ), which allows the preparation of strongly photonic systems (m  3.3) with negligible absorption in the red part of the visible spectrum. Using an etching technique, we prepared slabs of strongly scattering, ran- dom GaP networks with structural units of about 150 nm and two different porosities, 35 and 50%. Furthermore, the method makes it possible to control the thickness of the scatter- ing slab. The macroporous GaP structure is formed by anodic etching of n-type GaP single crys- tals (donor density, 2  10 17 cm -3 ) under dielectric breakdown conditions (7–9). The 350-m-thick polished wafers are mounted with a (100) face exposed to the 0.5 M H 2 SO 4 electrolyte. Application of a strong positive potential (15 V versus normal hydrogen elec- trode) leads to severe band bending at the n-GaP/electrolyte interface. Interband tunnel- ing of electrons (10) from the valence band or from band gap states to the conduction band takes place, generating holes at the surface, which are consumed in anodic dissolution of the GaP. The generation of holes is spatially nonuniform, which results in the porous net- work growing deeper into the GaP crystal. The anodic charge is proportional to the thickness of the porous structure L, enabling coulometric control of this important param- eter (9). A SEM micrograph of a cross section of the porous slab is used to calibrate the linear charge-thickness relation. The porosity is determined to be 35 volume % of air. We prepared a series of samples with porous slab thicknesses ranging from 5 to 120 m. In addition to the anodically etched GaP (A-GaP), we also prepared photoanodically etched GaP (PA-GaP) samples. Exploiting a technique that uses homogeneous photo-as- sisted etching, we are able to prepare slabs with a higher porosity. A-GaP samples are subjected to a further process of photoanodic etching in a H 2 O:H 2 SO 4 :H 2 O 2 electrolyte so- lution, using 50 mW of 1.96-eV sub– band gap light from a HeNe laser. Photons are absorbed by a transition of an electron from the top of the valence band to an interfacial state, 0.3 eV below the conduction band, followed by thermal release of the interfacial electron into the conduction band (11). The remaining hole is involved in anodic dissolu- tion. Because of the weak absorption of the light in the porous structure, the etch rate is 1 Van der Waals-Zeeman Instituut, Universiteit van Amsterdam, Valckenierstraat 65, 1018 XE Amster- dam, The Netherlands. 2 Debye Instituut, Universiteit Utrecht, Post Office Box 80000, 3508 TA Utrecht, The Netherlands. *To whom correspondence should be addressed. E- mail: schuurma@phys.uva.nl †Present address: Basic Sciences Center, National Re- newable Energy Laboratory, 1617 Cole Boulevard, Golden, CO 80401–3393, USA. R EPORTS www.sciencemag.org SCIENCE VOL 284 2 APRIL 1999 141