Reconfigurable Array Interconnection by Photorefractive Volume Holography

Parallel computing algorithms can be effectively implemented by combining local electronic processing with global optical interconnection. This dissertation describes the development of an optical array interconnection network based on photorefractive volume holography. The approach taken uses the correlation matrix-tensor multiplier (CMTM) algorithm, which optically convolves the phase-encoded input array with a control image holding the interconnection weights. Two-dimensional arrays can be interconnected with complex grey-level weights using binary phase-only spatial light modulators. The CMTM algorithm allows graceful accomodation of limited modulator size by trading off control image bandwidth for output signal to noise ratio. The optical correlation was performed by photorefractive four-wave mixing, storing the interconnection information in a single exposure of the control image. Multiple interconection patterns were prestored as color-multiplexed volume reflection holograms in z-cut LiNbO_3. Fast reconfiguration between interconnection patterns is possible using a wavelength tunable source, decoupling both the modulation and switching speeds from the slow photorefractive response. Experimental results confirmed theoretical predictions that the algorithm works best for densely connected networks, with a large fan-in to each output. Interconnection of up to 4096 inputs and outputs was demonstrated using such dense interconnection patterns. An aggregate average SNR of over 200 was obtained for 1024 inputs and outputs. Finally, a compact packaged optoelectonic processor system using CMTM interconnection is proposed, and its scaling behavior investigated.