25 July 2020 | New Scientist | 39describe the behaviour of electrons – to
solitons, lone travelling waves that share
mathematical descriptions with earthquakes
and stock market behaviour.
Nader Engheta at the University of
Pennsylvania and his colleagues have
shown that fast optical analogue computing
is possible using interactions between light
and matter. They have used a complex
structure known as a metamaterial to affect
the path of light in such a way that it can solve
integral equations. Their prototype, which
looks like a Swiss cheese about half a metre
or so across, is designed to work on the long
wavelengths of microwave radiation. Future
iterations could do so on optical or shorter
wavelength radiation, allowing the structure
to be a thousand times smaller and faster.Brain training
One particularly influential backer of
analogue computing has been the US
Defense Advanced Research Projects
Agency (DARPA), which invests in innovative
technology. In 2016, DARPA sought designs
for analogue or hybrid analogue/digital
devices that could provide the capabilities of
a supercomputer in a desktop device. Some
of the most promising ideas to come out of
that scheme revolved around electronic
devices called memristors.
Any time an electric current flows
through a circuit, it encounters resistance.
In a memristor, that resistance changes in
response to previous use, and the altered
resistance state is retained when the circuit
is off, meaning it has memory. This is useful
for computing as power-free data storage,
but also interests scientists working on
neuromorphic computing, in which
electronic circuits are used to mimic
the workings of a brain.
The strength of the connections, or
synapses, between the brain’s neurons
grows stronger as more signals pass through
them and weakens if signals become rare,
giving them learning functions akin to
muscle memory. This continuously varying
connectivity is awkward to replicate withequations are mathematical expressions in
which quantities are related in terms of
their rate of change rather than just their
values. The digital approach to tackling them
involves calculating and storing the value of
each point along a function relating two
variables, and then performing calculations
on those stored values. An analogue
computer, by contrast, would be able
to work on the whole function at once.
One way of doing this is to harness
the mathematics that governs electrical
circuits themselves. Quantities like electric
current, charge and capacitance are related
by rates of change in their values. This means
they fit differential equations, allowing
electrical circuits to serve as analogues
for all other systems governed by such
mathematical expressions. That is why
Tsividis and his colleagues have used
these kinds of circuit elements to develop
all-electrical analogue chips.
Unlike the analogue computers of the
1940s and 1950s, with their punch cards
and primitive wiring, these chips benefit
from all the same advances in semiconductor
research that have made digital computers
smaller and faster. These new analogue
chips can connect to each other as well as to
conventional digital computers, and – most
importantly – can solve certain problems
faster and more efficiently than their digital
counterparts. For example, multiplying two
eight-digit binary values digitally would take
about 3000 transistors, but an analogue
computer would need a maximum of eight.
“People did not seem to have considered
analogue computing in the context of
modern technology,” says Tsividis. “We
did, and things looked very promising.”
It isn’t only electronics that could be
useful for analogue computing. Researchers
are now turning to beams of light, not least
because of the possibility of superfast data
transfer. Optical technology offers other
benefits as well. When you put objects in the
way of light, for example, you get effects that
provide a physical analogue for a wide range
of phenomena. These range from simple
dispersion – which can also be used to
“ It would take about
3000 transistors to
do the calculation
digitally. An
analogue computer
would need eight”
>