106 MARCH2020|COMPUTERSHOPPER|ISSUE385
differential equations, an associated
digital computer can work out the
required connections and even make
those connections itself.Accuracy still
remains an issue,but advocates
suggest that some applications can
make do with limited accuracy.Soan
analogue processor could work
alongside atraditional digital processor,
with the ideal technology being
selected automatically foreachtask.
Alternativeelectronic
architectures
Another evolutionary concept is to
remain with conventional electronic
circuitry,perhaps even sticking with
good old-fashioned silicon, but to make
major changes to the architecture.
Whether new architectures will replace
current microprocessors or just sit
alongside themremains to be seen, but
most pundits are suggesting the latter,
at least in the short term.
This has started already,asGPUs
initially developed forimproving
graphics performance,especially in
gaming, are now being used as key
elements in high-performance
machines forscientific applications.
Another example is artificial neural
network (ANN) co-processors. ANNs
are one of the more revolutionary
paradigms and could, in theory,be
produced using analogue electronic
circuitry.Inthe short to mediumterm,
however,neural networks are being
implemented digitally in silicon chips
that work together with more
conventional microprocessors.
One of the more ground-breaking
potential new architectures is being
promoted to break the so-called von
Neumann bottleneck. One of the major
factors that throttles performance
in today’s processors, the von
Neumann bottleneckreferstothe
time-consuming requirement to move
data frommemory intothe CPU to be
processed, before shuttling it back
again afterwards. The revolutionary
concept that could offer some relief
here is to carry out processing within
the memory itself.
The quantum promise
We’ve been promised quantum
computing fordecades but, with Intel,
Microsoftand IBMnow on board with
active research projects, perhaps
mainstreamapplications are within
sight. It’s bizarre in the extreme –
being based on concepts such as
superposition, which means that abit
can represent both a0and a1atthe
same time –but theperformance
gains on offer are phenomenal.
With conventional computers, to
double the speed while keeping
everything else the same requires twice
the amount of hardware.With a
quantumcomputer,it’s necessary only
to addanextra bit, or qubit as it’s
called. So,for example,while the
much-heralded change from32 to 64
bits in mainstreamchips gave us a
welcome performance boost, in the
world of quantumdigital computers,
such achange would have resulted in
the speed increasing by afactor of four
billion. But all of this is fiendishly
difficult to achieve,since it requires the
circuitry to be held at temperatures
that are even lower than interstellar
space,while being totallyshielded from
external radio signals and all other
forms of interference.
While the general-purpose digital
quantumis probably afew years off
yet, it’s interesting to notethat
Canadian company D-Wave has had a
machine on the market forseveral
years. Used by Google and NASA,
this machine isn’t the quantum
equivalent of today’s digital
ABOVE:According
to IBM, in five
yearsquantum
computing will
go beyond the
research lab
and become
mainstream
