Integrated
optoelectronics
is
emerging
as
a
promising
platform
of
neural
network
accelerator,
which
affords
efficient
in-memory
computing
and
high
bandwidth
interconnectivity.
The
inherent
optoelectronic
noises,
however,
make
the
photonic
systems
error-prone
in
practice.
It
thus
imperative
to
devise
strategies
mitigate
and,
if
possible,
harness
noises
systems.
Here,
we
demonstrate
generative
part
adversarial
(GAN).
This
implemented
with
core
consisting
an
array
programable
phase-change
memory
cells
perform
four-element
vector-vector
dot
multiplication.
GAN
can
generate
handwritten
number
(“7”)
experiments
full
10
digits
simulation.
We
realize
optical
random
generator,
apply
noise-aware
training
by
injecting
additional
noise,
network’s
resilience
hardware
nonidealities.
Our
results
suggest
potential
more
complex
networks
based
on
large-scale,
realistic
hardware.
ACS Applied Materials & Interfaces,
Год журнала:
2020,
Номер
12(19), С. 21827 - 21836
Опубликована: Апрель 16, 2020
Progress
in
integrated
nanophotonics
has
enabled
large-scale
programmable
photonic
circuits
(PICs)
for
general-purpose
electronic-photonic
systems
on
a
chip.
Relying
the
weak,
volatile
thermo-optic
or
electro-optic
effects,
such
usually
exhibit
limited
reconfigurability
along
with
high
energy
consumption
and
large
footprints.
These
challenges
can
be
addressed
by
resorting
to
chalcogenide
phase-change
materials
(PCMs)
as
Ge2Sb2Te5
(GST)
that
provide
substantial
optical
contrast
self-holding
fashion
upon
phase
transitions.
However,
current
PCM-based
applications
are
single
devices
simple
PICs
due
poor
scalability
of
electrical
self-heating
actuation
approaches.
Thermal-conduction
heating
via
external
heaters,
instead,
allows
integration
large-area
switching,
but
fast
energy-efficient
control
is
yet
show.
Here,
we
model
switching
GST-clad
nanophotonic
structures
graphene
heaters
based
GST-on-silicon
platform.
Thanks
ultra-low
heat
capacity
in-plane
thermal
conductivity
graphene,
proposed
speed
~80
MHz
efficiency
19.2
aJ/nm^3
(6.6
aJ/nm^3)
crystallization
(amorphization)
while
achieving
complete
transitions
ensure
strong
attenuation
(~6.46
dB/micron)
(~0.28
dB/micron
at
1550
nm)
modulation.
Compared
indium
tin
oxide
silicon
p-i-n
display
two
orders
magnitude
higher
figure
merits
overall
performance.
Our
work
facilitates
analysis
understanding
thermal-conduction
heating-enabled
supports
development
future
systems.
Optica,
Год журнала:
2020,
Номер
7(3), С. 218 - 218
Опубликована: Янв. 22, 2020
Advances
in
artificial
intelligence
have
greatly
increased
demand
for
data-intensive
computing.
Integrated
photonics
is
a
promising
approach
to
meet
this
big-data
processing
due
its
potential
wide
bandwidth,
high
speed,
low
latency,
and
low-energy
Photonic
computing
using
phase-change
materials
combines
the
benefits
of
integrated
co-located
data
storage,
which
late
has
evolved
rapidly
as
an
emerging
area
interest.
In
spite
rapid
advances
demonstrations
field
on
both
silicon
nitride
platforms,
clear
pathway
towards
choosing
between
two
been
lacking.
paper,
we
systematically
evaluate
compare
computation
performance
platform
platform.
Our
experimental
results
show
that
while
platforms
are
superior
terms
integration,
modulation
device
footprint,
they
require
trade-offs
energy
efficiency.
We
then
successfully
demonstrate
single-pulse
optical
memory
photonic
waveguides
efficient
programming,
retention,
readout
>4
bits
per
cell.
paves
way
in-memory
Optics Letters,
Год журнала:
2021,
Номер
46(22), С. 5671 - 5671
Опубликована: Окт. 19, 2021
Programmable
photonic
integrated
circuits
are
emerging
as
an
attractive
platform
for
applications
such
quantum
information
processing
and
artificial
neural
networks.
However,
current
programmable
limited
in
scalability
by
the
lack
of
low-power
low-loss
phase
shifters
commercial
foundries.
Here,
we
demonstrate
a
compact
shifter
with
microelectromechanical
system
(MEMS)
actuation
on
silicon
photonics
foundry
(IMEC's
iSiPP50G).
The
device
attains
(2.9π±π)
shift
at
1550
nm,
insertion
loss
(0.33-0.10+0.15)dB,
Vπ
(10.7-1.4+2.2)V,
Lπ
(17.2-4.3+8.8)µm.
We
also
measured
bandwidth
f-3dB
1.03
MHz
air.
believe
that
our
demonstration
MEMS
implemented
compatible
technology
lifts
main
roadblock
toward
scale-up
circuits.
ACS Photonics,
Год журнала:
2022,
Номер
9(6), С. 2142 - 2150
Опубликована: Май 6, 2022
Programmable
photonic
integrated
circuits
(PICs)
have
recently
gained
significant
interest
because
of
their
potential
in
creating
next-generation
technologies
ranging
from
artificial
neural
networks
and
microwave
photonics
to
quantum
information
processing.
The
fundamental
building
block
such
programmable
PICs
is
a
2
×
unit,
traditionally
controlled
by
the
thermo-optic
or
free-carrier
dispersion.
However,
these
implementations
are
power-hungry
volatile
large
footprint
(typically
>100
μm).
Therefore,
truly
"set-and-forget"-type
unit
with
zero
static
power
consumption
highly
desirable
for
large-scale
PICs.
Here,
we
report
broadband
nonvolatile
electrically
silicon
based
on
phase-change
material
Ge2Sb2Te5.
directional
coupler-type
exhibits
compact
coupling
length
(64
μm),
small
insertion
loss
(∼2
dB),
minimal
crosstalk
(<−8
dB)
across
entire
telecommunication
C-band
while
maintaining
record-high
endurance
over
2800
switching
cycles
without
performance
degradation.
This
constitutes
critical
component
realizing
future
generic
systems.
Journal of Lightwave Technology,
Год журнала:
2021,
Номер
40(8), С. 2214 - 2243
Опубликована: Дек. 21, 2021
This
Tutorial
will
discuss
the
motivation,
benefits,
and
challenges
of
photonic
switching
in
data
centers
cover
prospects
future
involving
emerging
new
technologies
cross-layer
solutions.
The
primary
motivation
for
considering
rises
from
need
energy-efficient
scalable
intra-data
center
networks
to
meet
rapid
increases
traffic
driven
by
applications,
including
machine
learning.
inside
(East-West
traffic)
is
typically
significantly
greater
than
that
coming
out
(North-South
(Benson
et
al.
2010).
To
accommodate
such
traffic,
today's
large-scale
employ
cascaded
stages
many
power-hungry
electronic
packet
switches
interconnected
across
network
fixed
hierarchical
communication
topologies.
These
add
significant
latency
energy
consumption
while
limiting
bandwidth.
On
other
hand,
can,
principle,
support
interconnections
at
very
high
rates
on
parallel
wavelengths
keeping
their
nearly
independent
switch
port
Numerous
research
papers
have
predicted
benefits
scalability,
throughput,
power
efficiency
deploying
centers.
However,
not
yet
widely
deployed
commercial
warehouse-scale
time
writing
this
due
challenges.
They
are
related
1)
issues
control
management
planes
together
with
integrity
during
switching,
2)
scalability
>5000
racks
(>a
quarter-million
servers),
3)
performance
monitoring
required
reliable
operation,
4)
currently
existing
standards
allowing
limited
margin
(3
dB),
5)
practical
(technology-dependent)
relating
polarization
sensitivity,
temperature
cost,
etc.
In
telecom,
deployments
reconfigurable
optical
add-drop
multiplexers
(ROADMs)
(Lightwave,
2003)–(Perrin,
2015)
also
had
faced
similar
took
place
ten
years
after
first
testbed
demonstrations
1997
(Garrett,
fully
implemented
planes.
centers,
far
more
scale
dynamicity
traffic.
We
possible
solutions
methods,
topologies,
innovative
technologies.
particular,
broadly
surveys
state-of-the-art
technologies,
architectures,
experimental
results,
further
covers
details
arrayed-waveguide-grating-router-based
fabrics
offering
hybrid
methods
distributed
towards
networking.
ACS Photonics,
Год журнала:
2022,
Номер
9(10), С. 3181 - 3195
Опубликована: Сен. 29, 2022
Programmable
photonics
have
the
potential
to
completely
transform
a
range
of
emerging
applications,
including
optical
computing,
signal
processing,
light
detecting
and
ranging,
quantum
applications.
However,
implementing
energy-efficient
large-scale
systems
remains
elusive
because
commonly
used
programmable
photonic
approaches
are
volatile
energy-hungry.
Recent
results
on
nonvolatile
phase-change
material
(PCM)
integrated
present
promising
opportunity
create
truly
photonics.
The
ability
drastically
change
refractive
index
PCMs
in
fashion
allows
creating
units
with
zero-static
energy.
By
taking
advantage
electrical
control,
reconfiguration,
zero
crosstalk
between
each
unit,
can
enable
extra
(ELSI)
In
this
Perspective,
we
briefly
review
recent
progress
PCM
discuss
challenges
limitations
technology.
We
argue
that
energy
efficiency
is
more
critical
parameter
than
operating
speed
for
photonics,
making
an
ideal
candidate.
This
has
disruptive
paradigm
shift
reconfigurable
research
philosophy,
as
slow
but
large
modulation
provide
better
solution
ELSI
fast
power-hungry,
small
tuning
methods.
also
highlight
exciting
opportunities
leverage
wide
bandgap
visible-wavelength
such
optogenetics,
rewritable
circuits
(PICs)
using
nanosecond
pulsed
lasers.
latter
dramatically
reduce
fabrication
cost
PICs
democratize
PIC
manufacturing
process
rapid
prototyping.
Abstract
High-speed
optical
interconnects
of
data
centers
and
high
performance
computers
(HPC)
have
become
the
rapid
development
direction
in
field
communication
owing
to
explosive
growth
market
demand.
Currently,
interconnect
systems
are
moving
towards
higher
capacity
integration.
High-sensitivity
receivers
with
avalanche
photodiodes
(APDs)
paid
more
attention
due
capability
enhance
gain
bandwidth.
The
impact
ionization
coefficient
ratio
is
one
crucial
parameter
for
photodiode
optimization,
which
significantly
affects
excess
noise
bandwidth
product
(GBP).
silicon-germanium
(Si-Ge)
APDs
promising
thanks
low
silicon,
simple
structure,
CMOS
compatible
process.
Separate
absorption
charge
multiplication
(SACM)
structures
typically
adopted
Si-Ge
achieve
noise.
This
paper
reviews
design
optimization
high-speed
APDs,
including
advanced
APD
structures,
modeling
receivers.
ACS Photonics,
Год журнала:
2023,
Номер
10(7), С. 2020 - 2030
Опубликована: Фев. 16, 2023
To
keep
up
with
the
growing
bandwidth
demands,
photonic
integrated
circuits
(PICs)
have
been
widely
employed
in
various
application
scenarios
where
high
capacity
and
high-bandwidth-density
interconnects
are
required.
However,
it
is
challenging
to
scale
PICs
toward
future
petabit
per
second
requirements.
We
study
scalability
bottlenecks
of
terms
guiding
materials,
dense
integration
approaches,
wide-band
optical
sources,
high-efficiency
tunable
modulation
devices.
also
look
for
possible
solutions
address
these
challenges.
In
end,
we
provide
a
perspective
on
PIC
development.
Moore's
law
photonics
may
last
much
shorter
time
than
that
microelectronics
industry,
requiring
significant
innovations
technological
breakthroughs
research.
Nature Communications,
Год журнала:
2025,
Номер
16(1)
Опубликована: Янв. 2, 2025
Silicon
photonic
signal
processors
promise
a
new
generation
of
processing
hardware
with
significant
advancements
in
bandwidth,
low
power
consumption,
and
minimal
latency.
Programmable
silicon
processors,
facilitated
by
tuning
elements,
can
reduce
development
cycles
costs.
However,
traditional
programmable
based
on
optical
switches
face
scalability
performance
challenges
due
to
control
complexity
transmission
losses.
Here,
we
propose
scalable
parallel
processor
for
versatile
applications
interleaving
wavelength
temporal
dimensions.
Additionally,
it
incorporates
ultra-low-loss
waveguides
low-phase-error
switch
techniques,
achieving
an
overall
insertion
loss
10
dB.
This
design
offers
loss,
high
scalability,
simplified
control,
enabling
advanced
functionalities
such
as
accurate
microwave
reception,
narrowband
filtering,
wide-bandwidth
arbitrary
waveform
generation,
high-speed
computing
without
the
need
elements
calibration.
Our
demonstrates
advantages
both
scale
performance,
marking
advancement
large-scale,
high-performance,
multifunctional
systems.
work
presents
leveraging
With
switches,
be
programmed
enable
functions,
showcasing
great
Nanophotonics,
Год журнала:
2019,
Номер
9(1), С. 159 - 166
Опубликована: Ноя. 19, 2019
Abstract
Adiabatic
waveguide
taper
and
on-chip
wavelength
demultiplexer
are
the
key
components
of
photonic
integrated
circuits.
However,
these
two
kinds
devices
which
were
designed
by
traditional
semi-analytic
methods
or
brute-force
search
usually
have
large
size.
Here,
based
on
regularized
digital
metamaterials,
a
two-channel
focused
with
footprint
2.4
×
10
μm
2
has
been
proposed.
The
can
directly
connect
to
grating
coupler
under
absence
long
adiabatic
taper.
objective
first
method
modified
steepest
descent
used
design
splits
1520
nm
1580
light.
Experimental
results
show
that
insertion
loss
upper
(lower)
channel
is
−1.77
dB
(−2.10
dB)
crosstalk
−25.17
(−12.14
dB).
Besides,
simulation
indicate
fabrication
tolerance
device
reach
±20
in
etching
depth
±10
plane
size
changing.
Benefitted
from
extensibility
method,
other
types
ultra-compact
“focused”
devices,
like
mode
splitters,
converters,
power
splitters
also
be
designed.
Most
importantly,
this
more
complicated
functionalities,
such
as
multi-channel
demultiplexers.