Page 491 - Introduction to Information Optics

476 9. Computing with Optics

9.1. INTRODUCTION

The ever-increasing demand for high-speed computation and efficient pro-
cessing of large amounts of data in a variety of applications have motivated
the advancement of large-scale digital computers. Conventional computers
suffer from several communication problems such as the Von Neumann
interconnection bottleneck, limited bandwidth, and clock skew. The bandwidth
and clock skew limit processing speed and add design complexity to the
system. Communication bottlenecks at the architectural, bus, and chip levels
usually come from the utilization of time multiplexing to compensate for the
inability of electronics to implement huge interconnections in parallel. The
sequential nature of data transport and processing prevents us from building
high-performance computing systems. To overcome these bottlenecks, single-
instruction multiple-data (SIMD)-based and multiple-instruction multiple-
data (MI MD)- based architectures have been investigated. However, such
architectures still suffer from interconnection bottlenecks. Another promising
approach is to incorporate the attractive features of optics with those of digital
computers [1-7], yielding a hybrid optoelectronic computer system.
Optics has several advantages over electronics, including ultrahigh process-
ing speed, large bandwidth, massive parallelism, and noninterfering propaga-
tion of light rays. As a result, the application of optics has been widely
investigated for arithmetic and logical processing, image processing, neural
networks, data storage, interconnection networks, and ultrafast system bus.
From the viewpoint of data representation, currently available optical proces-
sors can be classified into two categories, analog and digital; from the
operational viewpoint, they are classified into numeric and nonnumeric pro-
cessors. In earlier periods, optics was used to process only analog signals.
However, in the last two decades, tremendous advances in nonlinear optoelec-
tronic logic and switching devices have invigorated the research in digital
optics.
Digital computing has excellent features including flexibility in computation,
easy implementation, minimal effects of noise, and lower requirement of devices
to identify two states of signal. Various optoelectronic systems have been
demonstrated, including the pipelined array processor using symmetric self-
electro-optic-effect devices (S-SEED) [8], optical cellular logic image proces-
sor [9], and the programmable digital optical computer using smart pixel
arrays [10,11]. Digital optical circuits can be constructed by cascading
two-dimensional (2D) planar arrays of logic gates using free-space interconnec-
tions. These programmable logic arrays can be used to implement various
complex functions, such as arithmetic and logical operations in constant time.
Recent advances in the algorithms, architectures, and optoelectronic systems
that exploit the benefits of optics have led to the development of high-