A short review of smart pixel technology is followed by an application of one specific smart pixel technology to a 2-D application of analog-to-digital conversion called digital image halftoning. A novel approach to digital image halftoning isdescribed based on a symmetric error diffusion algorithm, a new form of artificial neural network called an error diffusion neural network, and a smart pixel optoelectronic architecture. Two generations of smart pixel architectures are described thatincorporate self-electro-optic effect device (SEED) modulators flip-chip bonded to submicrometer feature size complementary metal-oxide semiconductor (CMOS) silicon circuitry to produce CMOS-SEED implementations of the error diffusion neural network.Simulations and experimental characterizations demonstrate that this hardware approach provides sufficient computational accuracy for the analog neural network while simultaneously providing switching speeds that support halftoning at video rates.
展开▼
