0%

Optical Computing

Optical Computing: Past and Future

Ravi Athale and Demetri Psaltis
Original
Website
,

Electronic computing’s historical strengths

Conventional wisdom says that electrons compute and photons communicate. That’s
because the strong Coulomb interactions of charged electrons can be leveraged
to perform nonlinear computations (Boolean logic), whereas charge-free photons
do not interact with each other at all in free space.

An electronic computer operates at baseband by manipulating the flow of charges
in semiconductors, such as silicon, whereas most optical systems transfer
information encoded on a carrier frequency of several hundred THz by the
polarization of bound electrons in dielectric materials, such as glasses.
Indeed, the lack of photon-photon interaction makes it possible to use a large
number of spatial and spectral channels to increase the information-carrying
capacity of optical communication systems.

Electronic telecommunication systems evolved from telegraphy, with baseband
operation using simple metal wires for transmission, through telephony, whose
increased demand for transmission bandwidth was initially met by higher carrier
frequencies that require more complex guiding structures such as coaxial
cables, and finally into fiber optic technology for long-haul communications
beginning in the late 1980s. The starting point for computers was similar to
that of early communications systems—that is, simple electrical circuits
operating at baseband frequencies, with bandwidth of several MHz.

Unlike communications however, computers have continued to operate at
baseband—and, thus far, with great success. Computers have acquired complexity
and speed through improvements in the resolution of lithography, which allowed
exponential gains under Moore’s law. Transistors with ever-shrinking dimensions
provide a highly localized interaction between electrical signals (typically
the gate and source voltages), through the 1/r2 drop of the electric field
established by the charge at the gate of the transistor. Such localization is
essential for Boolean logic, in which only two bits typically interact at a
time.

The miniaturization of transistors made possible by ever-finer-scale
lithography leads to increased speed, greater density, lower power and lower
cost (through increased integration)—all at the same time—and has formed the
basis of the microelectronic revolution. In contrast to communications, placing
the data on a high carrier frequency before performing nonlinear logic
computations has offered no obvious advantages.Electronic computing’s
historical strengths

Big data and physical limits

In recent years, it’s been widely recognized that conventional scaling in CMOS
processors is reaching its physical limits, and can’t provide the same
exponential improvement in computational capabilities as in the past. The
computational challenges now posed by so-called big-data analytics are also
necessitating a rethinking at a fundamental level.

All of this has driven increased interest in alternatives to silicon-CMOS-based
hardware for digital computation, a trend captured by a variety of campaigns by
industry groups to “reboot” information technology. These initiatives envision
tight integration among specific applications, alternative models of
computation, and new, potentially unconventional hardware platforms. Proposals
for building optical systems implementing a “reservoir” model of computation—a
variation on neural-net models—constitute one recent example.

In those efforts, nanophotonics could play a key role. The same advances in
lithography and manufacturing that have driven Moore’s law have also, in the
past decade, brought a veritable revolution in photonics technology, making it
possible to precisely create features far smaller than wavelength of light. As
a result, photonic-crystal structures, metamaterials, plasmonics and highly
resonant nanostructures are now enabling unprecedented control over light
propagation, modulation, generation and detection. Novel ideas in bottom-up
self-assembly of materials are also opening new vistas in light-matter
interactions through tools such as quantum dots.

These integrated-photonics developments are leading to exploration of
ever-smaller, ever-higher-performance devices for electrical-to-optical and
optical-to-electrical conversion. They have also rekindled interest in
nonlinear optical switches as logic devices for special-purpose digital optical
circuits—if not as CPUs in general-purpose digital computers. The rapidly
evolving landscape of information processing—and the increasing limits faced by
Moore’s law—makes now an opportune time to explore such advanced
optical-computing techniques.