►
From YouTube: AWG Meeting 013123
Description
Presentation by Akin Akturk of CoolCAD on "Silicon CMOS Operating at Cryogenic Temperatures?"
A
I
invited
Akin
who
I
met
to
Nest
on
our
Nano
fabrication
accelerator
program
and
he
he's
been
working
on
very
interesting
topics.
His
his
talk
right
now
or
today
is
going
to
be
silicon,
CMOS,
operating
acquisition,
temperatures
and
he's
gonna
address
or
discuss
the
recent
spike
in
interesting
cryogenic
applications
ranging
from
Quantum
Computing
to
read
out
integrated
circuits.
That
might
be
interesting
here.
A
The
other
reason
I
invited
him
is
because
he
he's
working
on
the
modeling
at
question
temperatures,
but
he
also
made
some
measurement
on
the
skywater
130
test
files.
So
he
has
done
some
work
in
there,
but
let
me
introduce
Dr
Akin
actor,
he's
the
founder
of
c
and
CTO
of
cool
cat,
which
is
located
at
College
Park
in
Maryland,
his
guest
researcher
at
nist,
and
he
got
his
PhD
in
2006
from
University
of
Maryland,
just
published
polymers
scientific
papers
on
silicon
and
silicon
carbide,
material
device
and
reliability
properties.
A
His
main
research
interests
in
in
Silicon
is
on
cryogenic
Silicon,
CMOS
measurements,
modeling
and
Design,
and
his
additional
research
interests
include
space
and
terrestrial
radiation
testing
and
modeling
and
Optical
sensor
designer
fabrication
and
testing.
So
thank
you.
Akin
for
joining
us
today
and
welcome
your
presentation.
B
A
B
Yep
I
can
see
it.
Okay,
so
I'll
minimize
this
out
of
the
way
and
these
to
the
slides.
Hopefully
it's.
A
B
Up
on
your
end,
perfect:
okay,
thank
you,
matey
for
the
introduction,
so
today,
I'm
going
to
talk
about
like
in
more
general
terms
like
the
the
physics
of
silicon
CMOS
operation
at
low
temperatures
and
how
we
model
them
and
also
talk
about
some
of
the
other
things
we
do
as
well.
So
as
Max
mentioned,
like
mine,
is
like
an
actor.
I
met
cool
cat
Electronics,
which
is
a
small
company
located
in
College
Park
Maryland.
B
We
spun
out
from
the
University
of
Maryland
I
used
to
work
at
universities,
but
I
don't
really
do
much
with
University.
At
the
moment,
I'm
affiliated
with
nist
to
I
live
literally
five
minutes
from
this
campus
and
I
I.
Do
lots
of
silicon
CMOS
measurements,
modeling
and
Design,
and
that's
how
we
started
and
we
got
the
name:
cool
cat
that
refers
to
low
temperature
cat,
but
over
time
be
switched
to
Second
carbide
more
because
we
used
to
do
that
as
well.
B
So
and
we
do
more
high
temperature,
modeling
and
measurements
now
and
but
that's
insane
carbide
and
when
it
comes
to
Silicon,
we
do
measurements,
modeling
and
design
for
silicon
cmoset
working
at
cryo.
So
next,
this
is
a
kind
of
repeats.
What
I
mentioned,
as
I
mentioned,
we
spell
from
the
University
of
Maryland
and
then
we
baking
like
the
second
half
of
two
thousands
and
it's
fun
by
me
and
my
ex
advisor
Neil
Neil
goldsman
and
we
have
lots
of
Publications.
But
the
armed
focused
areas
are
measureness,
modeling,
design
and
Fabrication
for
extreme
Electronics.
B
B
We
do
a
lot
of
work
on
seeing
carbide,
Power,
Electronics
beef,
I
design,
a
fabricate
our
own
power,
Moses
and
dials,
and
we
have
a
program
on
making
Optical
sensors
as
well,
usually
using
thin
carbide
for
the
ultraviolet
UVC
UVB
range
and
we
design
power,
circus,
DC,
DC,
DC,
AC
converters,
ranging
from
a
couple
hundred
watts
to
15
kilowatts,
and
we
are
a
member
of
various
organizations,
including
power,
America
and
I
work
with
this
all
the
time
and
we
work
with
all
the
almost
all
the
federal
agencies
and
a
lot
of
private
companies
as
well.
B
So
first
I
want
to
talk
like
a
few
slides,
but
I
want
to
show
a
few
slides
about
what
else
we
do
besides
silicon
low
temperature.
So
just
to
give
you
a
gist
of
like
what
we
are,
what
cool
kid
is
all
about.
We
do
lots
of
thick
and
carbide
fabrication,
so
these
are
some
of
the
the
devices
like
low
voltage,
high
voltage,
High
current
devices
we've
designed
and
fabricated.
B
So
these
are
all
our
our
own
designs,
like
the
we've
fabricated,
almost
all
of
them,
and
we
develop
lots
of
process
matters
for
the
fabrication
of
silicon
carbine.
The
same
is
like
we
can
fabricate
second
devices
too,
but
not
as
with
small
features
like
the
ones
you
can
get
from
skywater
or
in
other
places
we
use.
B
We
have
a
prototyping
Fab,
but
for
sting
carbide
we
also
are
they
have
our
own
processing
tools
for
very
high
temperature
processes,
so,
besides
fabrication
either
loss
of
radiation
testing,
maybe
they
can't
imagine
at
the
beginning,
New
Mexico
this
one
place,
I
go
go
for
this
test,
be
used
to
the
test,
silicon
CMOS,
as
well,
mostly
for
ionizing
those
radiation.
Using
a
couple
of
60
like
shown
here
and
besides
that
we
can't
we
can
do
all
the
radiation
effects.
B
You
can't
have
everyone's
like
at
the
terrestrial
level
and
also
in
space
for
ground
level.
We
measure
theater
in
time
curves
using
atmospheric
neutrons
at
Los
Alamos
for
space
radiation.
We
measure
total
ionizing
dose
or
have
ion
test
for
single
demand
effects
or
displacement
damage
effects
using
protons
and
the.
Lastly,
back
in
the
day,
we
developed
a
spice
look
called
cool
spice,
this
kind
of
on
the
back
burner
at
the
moment,
but
it
is
still
alive.
There's
a
version
on
the
web
for
students.
B
I
need
to
probably
update
the
and
put
the
new
version
on
the
web
page,
but
and
it's
initially
developed
to
model
low
temperature.
Second
CMOS:
that's
why
there's
cool
at
the
beginning,
like
in
front
of
spice
what
we
did
was
I'll
talk
about
it.
More
later
we
modified
the
existing
vision
for
equation,
set
and
Incorporated
some
low
temperature
physics
into
it
then
converted
the
whole
equation
into
a
c
code
and
compiled
it
cool
spice.
So
it's
not
based
on
behavioral
models.
B
It's
based
on
like
compiled
code
and
it
it's
like
a
module
spice
module
or
the
like
piecing
for
a
decent
tree.
But
more
recently
we
use
it
for
high
temperature
as
well,
and
we
Incorporated
the
thermal
Network
I
think
that's
actually
probably
needed
for
low
temperature
as
well,
so
for
the
power
devices,
self-heating
isn't
stuff,
Heating
and
heat.
Coupling
is
an
important
issue
and
I
think
salary
is
an
important
issue.
B
It
takes
some
temperatures
as
well,
so
you
can
overlay
a
electrical
network
with
a
thermal
Network
and
it
will
show
you
the
how
the
device
itself
yet
and
that
how
to
hit
couples
between
devices
that
are
adjacent
to
each
other
so
going.
So.
This
is
like
kind
of
the
other
things
we
do
so
I'm
gonna,
please
imagine
about
like
the
what
kind
of
low
temperature
capabilities
we
have.
B
So
we
have
low
temperature
probe
stations.
Also,
we
have
access
to
low
temperature
probe
stations,
they
are
usually
lecture
probe
stations,
as
shown
here.
These
are
open
cycle
systems.
So
if,
when
we
use
liquid
nitrogen,
we
can
go
down
to
77k
when
we
use
liquid
helium,
we
can
go
down
to
4K
and
we
also
have
a
crash
that
that
can
go
down
to
6K.
This
is
this
requires
the
fourth
packages.
B
Is
it
a
setup
for
that
at
the
moment
that
we
have
another
station
that
this
actually
can
go
low
or
high
temp?
B
So
the
in
in
the
high
temp
just
doesn't
go
as
low
as
the
other
ones,
but
it
can
go
up
to
six
seven
hundred
C
or
like
about
close
to
a
thousand
Kelvin,
and
we
can
prop
the
devices,
and
these
are
room
temperature
probe
station.
It's
initially
bought
for
making
high
voltage
measurements
so
here
that
the
Chuck
and
the
props
can
go
up
to
3000
volts,
and
so
it
uses
high
voltage
lines.
So
this
is
the
one
I
used
to
take
some
measurements
for
the
skywater
test
chip
made.
B
It
also
mentioned
earlier
so,
but
because
of
the
use
of
high
voltage
lines,
the
noise
floor
is
usually
higher
in
this
system,
because
it
it
either
is
the
high
voltage
piances
or
the
kit
that
tracks
cables
for
high
voltage,
but
the
same.
We
we
can
change
it.
We
both
new
equipment,
can
change
the
lines
to
the
regular
tracks
as
well,
so
the
next.
B
So
it's
like,
like
this
kind
of
repeats,
what
I
mentioned.
We
can
do
dciv
measurements
from
subteagrams,
using
standard
tracks,
cables
and
the
standard
Pro
clear,
prop
stations,
and
then,
but
if
you
do
long
integration,
we
can
decrease
the
noise
floor
to
some
extent
and
it
can
be
passed
or
non-pulse
or
it
can
be
some
arbitrary
waveforms.
B
We've
done
a
lot
of
AC
like
the
CD
and
transient
measurements
as
well.
Up
to
two
megahertz
and
our
CV
noise
floor
is
errands,
I
mean
they
changed
from
structure
structure
but
they're
on,
like
100
or
200
times
of
fat
and
the
one
thing
that
we
used
to
do
a
lot
and
I
don't
think
it's
standard
is
like
noise
measurements.
It's
either
one
over
F
or
RTD
noise
and
the
people
used
to
think
that
at
no
temperature
the
noise
will
go
away.
B
It's
usually
not
the
case,
so
these
are
some
examples
of
the
the
measurements
I
mentioned
before.
So
we
can
do
regular
mosfet,
like
ivcv
stuff,
as
shown
like
the
IVs
are
shown
here
or
if
you
wanna
do
like
a
gate.
Current
characterization,
we
usually
design
very
large
structures.
B
Then
we
can
measure
multiples
of
them
done
called
take
the
noise
floor
down
for
a
small
device.
Besides
standard
measurements,
we
we
use
the
a
lot
of
more
Niche
measurements.
For
example,
we've
been
characterizing
how
the
the
resistance
resistance
of
a
resistor
basically
changes
the
function
temperature.
This
is
to
determine
different
like
the
how
the
Dolphins
are
ionized.
So
that's
one
way
to
find
the
ionization
level
as
a
function
of
temperature.
It's
also
another
way
of
finding
ionization
as
a
function
of
field
and
or
current.
B
So
this
talks
more
about
the
CV
measurement
so
before
it
was
more
mostly
DC,
the
CV
for
the
CV
measurements.
We
use
a
Precision
LCR
meter
in
usually,
and
these
are
used
in
spice
model
extraction
and
finding
like
oxide
thicknesses
or
overlap,
Caps
or
Fringe
caps,
and
the
ring
oscillators
and
other
circuits
are
used
for
splice
model
verification.
B
So
I
think
this
is
my
last
slide
about
the
types
of
measurements
we
oh,
the
most
important
thing
is
the
noise.
So
that's
something
that's
usually
not
done
so
we
used
to
do
a
lot
of
modeling
and
low
temperature
extraction
for
people,
designing
infrared
sanssters
or
a
readout
circuits.
So
these
are
not.
B
These
are
usually
at
higher
temperatures,
not
at
4K,
but
one
thing
they
do
care.
A
lot
is
the
one
or
F
noise.
So
we've
been
doing
a
lot
of
fun
aware
of
noise
measurements,
so
the
for
that
one.
We
have
a
system
like
simplified.
The
simplified
version
of
it
is
shown
at
top
left
different,
like
low
noise,
amplifiers
and
Digital
Signal
analyzers
the
down.
We
take
the
free
Transformer,
the
power
spectral.
The
answer
of
this
measured
functions
and
determine
one
over
F
noise.
B
At
various
temperatures,
as
I
mentioned,
the
Iran
room
temperature,
the
the
convention
is
that
as
you
cool
down,
otherwise
the
noise
will
go
down,
there's
the
tasks
which
were
around
town,
but
if
we
keep
going
down
in
temperature
that
that
cancer
doesn't
change
Trend,
so
you
have
to
be
careful
about
the
the
noise
one
over
F
noise
and
for
a
if
you're,
designing
a
readout
integrated
circuit.
B
You
need
to
be
even
more
careful
about
the
next
thing
we
measure,
which
is
the
RTD
random,
Telegraph
noise.
So
these
are
pulses.
Then
they
there
can
be
like
two
different
sources
like
Origins
for
this
process
it
can
be
like
the
either
the
trapping
or
like
mobile
influxations,
depending
on
which
model
is
what
or
the
the
hugis
model.
B
But
the
the
idea
is.
This
is
a
lot
of
used
to
study
this
in
the
past,
not
that
low
temperature
at
room
temperature.
One
thing
they
say
is
like
once
you
trip
in
like
a
carrier,
then
you
change
the
the
potential
well
around
that
carrier,
so
it
takes
longer
to
detrap
it.
So,
but
these
are
very
long
time
scales
so
and
it's
very
tedious,
especially
doing
it
at
low
temperature
using
a
probe
station,
but
we
have
if
we
establish
a
very
low
noise
floor.
B
So
these
are
the
longest
measurements
we
ever
taken
lasted
as
long
as
10-15
minutes.
So
these
are
charging
and
discharging
taking
place
in
like
in
seconds
or
tons
of
seconds,
even
hundreds
of
seconds.
B
So,
if
you
ever
read
that
circuit
it
and
you
want
it
to
work
at
low
temp,
you
have
to
be
careful
about
this
RTD
rtn
noise,
because
the
one
is
going
to
go
to
low
temperatures
to
suppress
noise,
but
the
even
at
very,
very
low
temperatures
depending
on
the
technology.
You
can
get
there
these
pulses,
which
will
cause
problems
for
your
last
circuit,
so
I'm
gonna
mention
like
the
some
of
the
important
low
temperature
physics
like
what's
the
importance
or
second
CMOS
operating
at
low
temperature
and
how
we
can
model
and
capture
Its
Behavior.
B
So,
first
of
all,
I
think
this
is
the
Redundant
slide
for
this
crowd,
but
for
most
people
I
usually
talk
to
they.
They
don't
always
know
or
think
that
similar
device
would
work
at
very
low
temperatures.
And
if
you
look
at
the
very
old,
like
the
books
on
Sig
and
CMOS,
there's
a
reason
why
they
may
think
that
way.
B
So
because
in
the
in
the
past,
the
long
Channel
devices
and
because
of
the
processing,
you
won't
get
a
uniform
Channel
and
you
were
getting
like
pools
of
carriers
and
they
were
always
making
it
the
complete
channel
from
source
to
drain,
and
you
were
getting
not
ideal
curves
but
was
most
modern
device
to
work
at
low
temperature
and
the
IV
occurs.
B
B
So
we
in
the
past,
we
measured
lots
of
different
devices,
including
battle
core
SOI,
and
then
we
used
to
do
ticket
modeling
as
well,
and
this
shows
the
results
for
one
of
the
tkn
models
that
you
can
have
ionization
in
the
source
and
drain
regions
of
a
mosfet,
as
well
as
in
the
channel
in
depletion,
regions
around
Junctions
and
I'll.
Talk
about
why
that
is
the
case.
So
this
is
the
main
reason
normal
operating
devices
at
extreme
temperatures,
even
at
a
fraction
of
a
k.
B
So
what
happens
at
low
temperature
in
general
in
the
the
saturation
part
of
a
mosfet
is
to
go
lower
in
temp
you
get
higher
higher
currents.
That
is
usually
true,
but
if
you
go
to
lower
vgs
values,
that
will
that
may
or
may
not
be
true,
there's
a
crossover
point
below
which
the
the
current
will
go
down
with
temperature
decreasing,
but
in
general,
at
high
high
vgs
ibds,
the
higher
lower
the
temperature
you
get
more
current.
B
So
that's
why
you
get
usually
faster,
like
plugs
or
faster
operating
circuits
at
lower
temperatures,
but
depending
on
the
application
unit.
Another
reason
people
go
to
low
temperature,
so
the
running
a
circuit
at
a
faster
speed
is
one
reason
and
the
quantum
Computing
or
readout
circuits
for
images
is
another
reason,
but
you
also
need
to
care
be
careful
about
the
another
reason
people
got
low.
B
So
so
the
one
there
are
like
four
I
would
say:
there
are
four
important
physics
taking
place
at
door
temperature.
The
first
one
is
incomplete
ionization.
So
if
you
have,
if
none
of
the
the
impurities
ionized,
you
wouldn't
get
a
a
normal
function
in
most
with
it.
So
since
we
measure
standard
IV
curves,
that
means
something
is
working
fine.
So
that
means
the
source
train
has
to
work
and
also
child
has
to
work
so
what's
happening
at
the
source
and
rain,
so
source
and
drain.
As
you
guys
all
know,
it's
a
highly
top
regions.
B
So
when
you
have
a
fair
height
of
region,
you
have,
you
have
to
open
importance,
interact
with
each
other
and
they
form
their
own
band.
That's
called
imprint
event,
so
the
imprinted
band
levels
May
overlap
with
the
conduction
band
of
a
silicon
device.
When
that
happens,
so
the
top
the
the
cares
next
to
the
Dolphins
are
not
trapped,
so
they
can
easily
switch
between
the
open
devil
and
the
second
conduction
band
or
the
the
valence
band.
So
that
means,
if
you
drop
the
source
standard
rate
high
enough
in
the
pan
of
the
the
temperature.
B
You
always
have
lots
of
carriers.
This
has
been
done,
like
analyzed
many
many
years
ago
when
they
start
working
working
on
second,
so
the
earlier
papers
are
actually
from
like
the
50s
or
the
60s
and
then
so
that
explains
what
happens
in
The,
Source
Center
and
then
what
happens
in
the
the
channel.
So
one
reason
the
channel
is
ionized
or
the
depiction
regions
ionized
like
the
bottom
of
the
drain
and
bottom
of
the
sources.
B
You
have
high
electric
field,
it's
either
building
or
an
applied.
So
if
you
have
a
high
field,
you
have
a
band
binding,
and
if
you
have
a
trap
charge
with
the
the
impaired
like
the
Trap
level
or
the
dopen
level,
then
you
can
have
telling
all
right.
You're
not
gonna,
have
too
much
thermionic
darling,
but
you
can
also
have
Quantum
powering
so
that
that
the
final
effect
is
that
will
free
up
the
the
carry
there
so
then
give
an
ionized
opened.
B
So
there's
another
way
to
have
ionization
in
the
channel.
But
besides
these
two
like
the
either
you
have
high
imperative
High
development
levels
or
high
Fields.
There
are
also
a
couple
other
ways
to
ionize
the
opens
as
well.
It's
very
low
temperatures
I'm
not
going
to
go
into
detail
here,
but
you
can
have
to
open
the
impact.
The
insulation,
just
like
you,
have
a
regular
impact
isolation
from
band
to
band.
B
From
balance
the
conduction
band,
you
can
navigate
the
carriers
in
the
conduction
band
ionizing,
for
example,
donor
levels
that
are
about
40
to
50
me
mediev
below
the
the
conduction
band
level.
So
you
can
also
have
so
you
can
have
filter
installation,
you
can
have
current
asset
italianization
or
you
can
have
the
document
conduction
band
or
the
the
acceptor
valence
band
ionization.
B
So
if
we
used
to
do
lots
of
ticket
modeling,
so
there's
like
a
distributed
more
languages
or
distribution
equations
in
general,
so
these
are
coupled
semiconductor
equations.
So
it
includes
personal
equation
along
and
also
they're
gonna
hold
current
content
equation.
So
there
are
some
models
in
there
that
captures
the
incompletation.
B
We
wrote
the
paper
back
in
the
day
on
this,
we've
been
considering
three
different
models.
The
first
model
uses
just
the
standard
firmware
statistics,
so,
instead
of
like
EF,
you
convert
it
to
like
the
care,
concentration
and
also
activation
energy,
and
you
include
the
generous
levels.
So
that
tells
you,
if
you're
the
formula
was
low
enough
or
high
enough.
That
will
give
rise
the
higher
or
lower
ionization
by
itself.
B
So
there's
a
special
takes
place
at
hard,
open
levels,
so
the
the
hard
open
level
you're
you're,
more
likely
of
clusters
of
higher
Doppler
regions,
which
gives
rise
the
in
excess
or
industry
or
which
gives
rise
to
ionization.
B
So
one
thing
I
want
to
show
here.
This
is
not
for
silicon,
but
we've
been
doing
the
same
stuff
in
seeking
carbide
as
well.
Cigarette
low
temperature
is
similar
to
seeing
carbide
at
high
temperature,
so
you
get
the
same
almost
the
same
effects
so
the
because
they
in
this
case
the
activation
engines
are
higher
and
then
that
and
that
that
gives
a
rise
to
reduce
temperature,
reduce
ionization
and
so
the
room
temperature.
Seeing
carbide
is
kind
of
equivalent
to
very
low
temperatures.
B
B
You
attentive
high
ionization
too,
because
the
the
absence
of
carriers
I
give
rise
to
the
installation
at
the
dopping
level.
This
is
related
to
just
the
firmative
statistics
and
usually
there's
a
sweet
spot
in
between
around
like
around
10
or
18,
and
some
of
that
corresponds
the
the
more
transition
level.
So
it
is
not
high
the
concentrate
enough
or
low
capacitrate
enough,
so
they
don't
really
the
Dolphins
don't
talk
to
each
other,
but
they
partially
overlap.
B
So
that
task
to
give
rise
to
the
lower
ionization
allowance,
but
this
ionization
levels
assume
there's
no
like
the
field
and
also
there's
no
current
flowing.
So
if
you
have
this
additional
effects,
then
you
it
will
give
rise
the
higher
education.
So
that's
one
important
thing
happening
at
very
low
temperatures
and
so
the
what's
the
effect
of
that
on
the
the
mosfetime
weaker
one
place,
you
can
observe,
observe
the
the
change
in
generalization
with
field
and
or
current
is
the
like,
the
sub
threshold
or
the
weekend
version
region.
B
B
It
can
differ
from
what
you
would
observe
at
room
temperature,
but
you
need
to
be
careful
because
sometimes
in
this
part
of
the
curve,
you
can
also
see
that
the
president
mosfet
the
current
flow
on
the
side
was
at
the
end
of
the
width
so
contributing
so
you
can
have
it
like
a
double
hump.
B
B
So,
as
you
can
see
at
the
bottom
left
at
4K,
the
impact
organization
rates
rates
for
holes
and
electrons
are
much
higher
than
than
compared
to
the
room
temperature
value,
so
yeah
a
lot
more
minor,
the
carriers
so
that
that
can
have
a
huge
effect,
especially
if
you're
using
an
SOI
device.
So
the
those
minority
cares
has
to
go
somewhere
after
they
are
generate
or
created.
So
in
the
case
of
nsoi,
they
can
go
to
the
bottom
of
the
the
gate,
oxide
and
then
alter
threshold
voltage
locally
and
give
rise
the
current
Kings.
B
So
the
same
thing
can
happen
at
room
temperature,
but
it's
more
likely
to
have
that
long
time
and
even
for
bug
device,
it
can
give
rise
to
some
current
Kinks
if
you
don't
have
a
good
subset
contact
or
body
contact
and
and
other
thing
about
this,
as
I
mentioned,
is
the
current
Kings
and
so
another
effect
is
the.
B
This
is
not
really
a,
and
this
is
a
numerical
problem
if
you
do
a
t-can,
so
the
internet
concentration
decreases
significantly
at
lower
temperature
and
it
actually
loses
its
physical
meaning
at
some
point.
But
if
you
run
a
tcat,
very
low
in
high
values
or
in
terms
of
care
about,
this
can
give
rise
the
loss
of
numerical
errors,
so
that
makes
tick
installations
extremely
challenging
at
long
time.
But
that's
this
is
not
a
problem.
This
is
not
much
of
a
problem
for
compact
models.
B
You
can
get
around
it,
so
this
is
just
I'm
gonna
skip
this
one,
but
what
this
says
is
there
in
SOI,
if
you
create
the
minor
care,
so
they
can
go
to
the
bottom
of
the
to
the
next
to
the
black
kit
and
alter
the
potential
distribution
in
the
channel,
and
that
can
give
us
the
current
King.
So
these
are
examples
of
current
Kings,
actually
I.
B
These
two
I
believe
right
about
twice.
Then
this
may
be
NSO.
I
think
this
is
an
SOI
device,
so
you
can
even
see
that
Embark
transistors
as
well,
so
the
the
so
the
there's
something
that
needs
to
be
like
captured
by
the
spice
model
if
it
is
a
big
effect,
so
the
as
you
go
lower
and
voltages
than
obviously
you're
less
likely
to
have
impact
than
translation.
So
there's
there's
a
way
to
to
stop
press
it
by
using
lower
voltages,
but
then
the
lower
voltages.
B
So
so
far
we
discussed
two
effects:
one
of
the
incoming
realization,
the
other
one
is
the
impact
ionization.
The
third
effect
you
can
see
for
some
device
that
low
temperatures
is
the
short
Channel
barrier.
So,
instead
of
getting
a
linear
curve
in
the
like
a
in
the
linear
region
of
a
mosfet,
you
get
this
dark
Behavior.
That's
because
between
the
source
and
the
channel,
you
have
a
small
potential
barrier
and
if
the
that
barrier,
depending
on
how
the
mosfet
is
done,
it
may
show
up
in
the
IV
curves
at
very
low
temperatures.
B
So
this
is
actually
how
much
harder
to
capture
in
a
regular
spice
model
and
it
may
be
captured
using
a
like
a
sub-circuit
in
a
table,
some
kind
of
a
like
a
very
low
barrier,
diet
and
seriously
the
mosfet.
But
there's
some-
and
this
is
not
a
universal
thing.
B
None
of
the
other
stuff
is
universal,
either
so
depending
on
who
makes
it
but
which
technology
it
is
like
how
it's
made
you
either
see
all
these
effects,
or
you
see
one
or
two
of
them,
and
the
last
important
episode
that
there
were
three
effects
that
this
is
the
fourth
one
that
anybody
designing
low
temperature
circuits
needs
to
be
aware
of
is
the
the
mismatch
at
low
temperatures,
so
this
tends
to
be
a
bigger
problem.
B
We
use
that
many
years
ago,
but
I
think
there's
like
a
recently
a
paper
published
on
this
as
well,
so
for
usually
for
the
Smalls
with
unlocked
devices
or
Narrows
and
the
shortest
Channel
devices.
If
you
keep
measuring
different
device
on
different
dies
and
like
at
different
locations,
you'll
see
that
the
sub
threshold
part
can
be
all
over
the
place.
This
is
due
to
a
mismatch,
so
in
general,
I
do
recommend
not
using
sub
threshold
region
as
much
as
possible,
mainly
due
to
this
effect.
B
But
if
you
use
a,
if
you
don't
use
the
smallest
Channel
like
device,
it
usually
helps
with
this
effect.
They
tend
to
match
better
as
to
go.
I
mean
it.
It's
like
it's
an
expected
result,
so
next
I
hope
I'm
not
like
we're
running
through
it
really
fast,
but
the
next
I'm
going
to
talk
about
the
the
modeling,
so
we
used
to
the
tcat.
But
then
we
switched
we've
been
doing
more
spice,
more
link
or
compact,
more
link
to
just
show
the
ticket
I'll
skip
that,
but
it's
a
drift
division
equations.
B
B
So
these
are
some
of
the
the
ticket
simulations
I
believe
we
did
in
the
past
dot
20K
and
below
that
we
were
having
lots
of
numerical
problems,
data
the
Intrinsic
Care
consultation,
but
in
this
case
we
use
r
on
like
in-house
tcat,
the
one
we
wrote
and
it
uses
the
it's
a
regular
Chevrolet,
gamma
discretization
and
all
the
numerical
tricks
you
can
have
but
I
believe.
If
you
use
a
commercial
tool
like
like
from
civac
or
synopsis
or
some
other
place,
probably
it
can
handle
those
convergence
issues
better.
B
Obviously
you
cannot
use
the
ticket
model
for
the
to
get
like
modeling
for
the
whole
circuit,
so
you
have
to
have
a
compact
model
to
be
able
to
simulate
medium
to
large
circuits,
even
like
circuits
with
fear
of
mosfets,
so
we
have,
and
the
self-feeding
can
be
an
issue
which
I
I'm
not
going
to
talk
too
much
here,
but
the
diet
and
electrical
Network
overlapping
the
terminal
Network
as
well,
especially
at
very,
very
low
temperatures,
because
a
Kappa,
the
term
division
counts
and
tests
the
the
drop
at
the
very
low
temperature,
so
that
will
give
rise
to
a
very
high
thermal
resistance
that
gives
rise
to
the
level
of
self
eating.
B
So
back
in
the
day,
we
sometimes
not.
We
don't
do
it
as
much.
Now
it's
like
we
used
to
get
a
compact
model,
we
start
from
scratch.
Basically,
we
designed
we
used
to
design
a
circus.
These
are
some
of
the
stuff
we
did
over
the
years,
and
these
are
all
in
Silicon,
the
the
layouts
and
things
like
that
and
different
processes.
These
are
the
whole
processes
we
use
in
the
past,
they're
like
nearer
skywater
ones
as
well.
B
So
once
we
have
the
different
bit
length,
mosfets
different,
most
cap
structures,
different
dice
structure,
different
resistor
structures
like
the
contact,
nvs
structures,
we
get
them
fabricated
and
that
we
measure
them
to
extract
models.
So
these
are
some
of
the
additional
circuits
with
the.
So
how
is
the
compact
mode
link?
Traditionally
we've
been
doing
3D
doing
it
in
three
different
ways.
The
first
is,
as
I
mentioned,
we
have
our
own
spy,
so
we
modified
bc4
and
then
converted
into
C
code,
then
compile
it
in.
B
B
B
Okay,
so
that
it
can
be
around
much
as
an
equal
Supply
that
can
be
is
run
in
in
other
commercial
simulator
and
the
third
method,
which
is
like
the
usually
the
most
favored
method,
is
the
running
just
standard
piece
for
or
bcentury,
because
the
first,
the
first
method
is
obviously
it's
just
for
us
one,
mainly
unless
you
have
cruise.
But
it's.
The
second
method
is
like.
It
is
extremely
difficult
to
run
very
large
circuits,
using
just
real
game
models
and
about
the
third
model.
B
We
start
with
modifying
the
Beason
form
model
because
in
in
our
original,
like
trials,
the
standard
reason
for
wasn't
working
well
for
us,
but
the
I
believe
over
time.
We
learn
more,
and
so
we
got
better
at
feeding
these
and
four
models
to
measured
curves.
But
this
shows
like
if
you
we've
been
also
using
ICT,
which
is
a
commercial
software.
B
You
can
use
to
extract
some
spice
models,
so
we've
been
using
that
to
we've,
been
feeding
the
measured
data
and
extracting
bc4
models
using
that
at
different
temperatures.
But
it
wasn't
working
well,
so
that's
why
we
originally
modified
PS4.
B
Basically
modifying
means,
like
we
added
a
few
more
equations
capture,
the
the
four
effects
I
mentioned
earlier
and
you
don't
always
need
to
like
activate
all
four
like
equations
as
corresponding
those
all
all
those
four
like
the
the
physics,
but
depending
on
what
technology
you
model
you
can
get
like
activate
one
or
the
other.
B
So
with
that
modification
everything
was
working
much
much
better,
and
then
we
were
capturing.
The
we've
been
capturing,
the
the
freeze
of
artifacts
like
in
in
the
sub
thresholds
and
then
using
one
model
set
basically
for
different
WL
devices.
So
we've
been
always
almost
always
using
scalable
models,
because
that
makes
life
usually
easier
for
the
user,
but
we've
been,
we
haven't
been
doing
that
much
pinning.
B
But
if
more
accuracies
started
at
each
WL
device,
the
beginning
can
be
also
used
to
increase
the
accuracy
of
piston
for
a
teach
node,
and
so
we've
been
using
basically
one
model
card
for
various
temperatures
and
various
bits
and
legs,
even
though
only
one
with
unlocked
is
shown
here
and
that
things
usually
work
well
with
the
modified
piston
form
model.
B
And
but
then
we
started
getting
better
at
pc4
as
well,
then
just
started
using
that
the
besides,
that
the
other
reason
was
that
one
of
the
customers
we
worked
with,
they
had
a
circus
with
like
millions
of
transistors
in
them,
and
the
velocation
models
were
not
working,
so
they
were
having
lots
of
convergence
issues
there.
So
we
wanted
to
come,
go
back
to
the
the
star
business
for
because
the
most
of
the
commercial
submitters
are
optimized
for
for
the
standard
model,
so
they
they've
an
easy
time
running
large
circles
with
the
standard
models.
A
B
From
last
slide,
hello,
does
anyone
have
a
question?
I
mean
the.
So
these
are
the
next
slide
kind
of
summarizes.
What
we
do
we
can
do
low
temperature
measurements
and,
depending
on
the
system
we
use,
it
can
go
down
to
like
stop
hit
clients
in
terms
of
current,
and
then
we
can
do
dciv
CV.
B
Then
one
thing
we
used
to
do
a
lot
is
the
noise
measurements
either
one
or
F,
or
a
random
player
of
noise,
and
we've
been
working
with
Founders,
most
notably
like
skywater,
and
we
used
to
work
with
power
just
to
and
and
most
of
our
work
used
to
being
at
for
extracting
spice
models
without
had
77k,
because
that's
where
most
of
the
radar
circuit
designers
like
to
like
to
stay
or
if
that's
the
temperature,
they
are
comfortable
with,
because
everything
works.
B
The
the
imager
area
is
also
work
well
at
the
temperature
there
are
some
medical
applications
there
on
that
temperature
as
well,
so,
for
example,
for
scanners
and
things
like
that,
but
the
more
recently
there's
a
lot
of
interest
at
Quantum
Computing.
So
for
that
one,
obviously
most
people
want
to
be
at
like
4K
or
even
at
a
fraction
of
a
k,
so
but
all
the
equation
and
physics
that
are
still
valid,
even
at
very
low
temperatures.
B
But
what
happens
is
like
the
if
you,
for
example,
the
The
Source
Channel
barrier,
you
may
see
more
of
it
as
you
go
further
down
in
temperature
or
the
or
the
impact
ionization
will
just
we'll,
definitely
get
worse
as
you
go
lower
in
temperature,
whether
they
are
raised
to
mitigate
those
problems
or
work
around
it.
One
is
the
to
capture
in
model
all
these
like
subtleties
and
otherwise
to
choose
a
technology
that
can
get
around
it.
B
For
example,
as
far
as
impact
organization
is
concerned,
it
is
better
to
use
a
bug
technology
than
an
SOI,
but
sometimes
they
may
or
may
I
mean.
There
are
also
reasons
to
be.
Some
people
want
to
use
it
as
well,
but
as
far
as
like
the
most
valid
vehicles
are
concerned,
they
are
more
standard.
B
Looking
if
this
about
the
wise
and
also
advanced
technology
is
chosen
like
the
but
as
long
as
there's
like
enough
treasure
voltage
offs
like
the
Headroom,
it
would
be
good
to
decrease
the,
for
example,
the
voltages
to
get
around
the
impact
analyzation
problem
as
well.
B
It's
just
there's
too
much
work
here,
but
this
kind
of
is
what
I
mentioned
before
we
can
measure
and
model
devices
from
4K
up
to
900
or
1000k,
using
different
prop
stations,
because
the
low
temperature
probe
stations
can
go
from
4K
to
room
temp
and
there's
another
one,
because
it
goes
from
from
temp
up
to
700c
and
and
so
far
our
most
credit
work
has
been
focused
on
operation
is
77k,
but
the
emergence
of
quantum
Computing
4K
seems
to
be
like
up
and
coming
more
people
are
interested
in
4k
operation.
B
Then
they
use
different.
They
they
didn't
they
weren't
as
interested
back
in
the
day.
So
we
work
with.
In
the
past,
we
worked
with
skywater
to
model
their
Knight
and
limited
process,
but
that's
part
of
this
Google
collaboration,
which
mate
is
also
part
of
the
or
you
guys,
are
also
part
of
the
one
13
millimeters,
the
the
technology
of
choice
and
then
what
we
did
before
for
Sky,
whatever,
which
I
think
is
a
good
way
of
modeling.
B
Everything
is
much
more
likely
to
work
like
if
you
have
a
very
large
circuit,
you
don't
need
to
worry
about
whether
you'll
get
an
answer
or
not.
You'll
eventually
get
an
answer
because
and
you
get
an
answer
and
a
reasonable
amount
of
time,
and
but
if
a
lot
of
accuracy
is
desired
at
a
certain
device
level,
the
bint
version
of
the
bsn4
model
can
be
used.
The
bins
version
means
in
the
so
normal
for
the
B7
for
scalable
model.
B
You
choose
it
with
a
length
scale,
and
you
said
my
model
is
built
in
this
like
the
before
between
this
element
and
L
Max
and
wb9w
Max,
but
instead
what
you
can
do
is
like.
You
can
divide
that
region
into
smaller
rectangles,
so
you
can
have
multiple
L's
and
multiple
W's,
and
if
you
have
like
one
addition,
I
wanted
to
show
W.
That
means
you
have
four
regions
for
the
whole
so
that
that
extends
the
whole
wnl
span.
B
So
then,
you
have
different
models
and
this
the
commercial
Spicer,
the
simulators
usually
can
seamlessly
switch
between
different
models.
But
you
need
to
be
careful
about
the
bin
model.
It
needs
to
be
continuous
along
the
edges
and
at
the
corners,
so
there's
a
there's
a
like
some
kind
of
art
doing
the
beginning,
so
you
cannot
just
arbitrarily
choose
a
different
spice
model
for
each
Bean,
so
it
needs
to
be
a
continuous
along
the
edges.
So
there
are
three
ways:
we
approach
modeling
low
temperatures.
B
B
B
That
way
we
achieve
speed
and
also
achieve
the
accuracy
and
but
that
wasn't
good
for
everybody,
because
then
the
initiative
of
our
spice
and
it's,
which
is
its
own
limits.
It's
not
like
a
kitten
Spectra.
So
the
second
method
was
like
What.
If
you
wanna
use
those
equation
steps,
but
in
your
simulator,
like
whatever
you
use,
it
can
be,
for
example,
Spectra
or
a
spice,
so
in
that
case
we
convert
it
into
the
barrel.
B
K
abilities
captures
all
the
equations
all
the
modifications
of
the
equations,
but
the
the
problem
with
that
was
for
very
large
circles.
We've
been
having
like
convergence
issues
for
like
small,
the
medium
size
circuits.
It
was
everything
was
fine,
so
the
fan
has
gone.
If
you
want
to
use
like
millions
of
chances
in
a
circuit,
then
it
may
not
work.
But
if
you
want
to
use
like
100
transistors
in
a
circuit,
then
it's
probably
fine
and
the
third
version
is
just
using
the
standard
PCM
and
leave
it
as
like
the
the
pros
and
cons.
B
So
the
the
cons
is
like
it
may
not
capture
every
single
detail,
but
that
the
pro
is
like
you
can't
simulate
you're
a
lot
less
active
in
the
convergence
problems
and
you
can't
get
an
answer
even
for
very
large
circuits,
so
I
think
I.
Okay.
This
is
what
this
was
like.
The
last
thing
I
wanted
to
mention.
A
Okay
thanks
so
much
Akin,
I,
guess
we'll
open
up
for
questions
and
I.
Think
Professor
Christian
has
a
question
so.
A
So
what
what
Christian
is
fixing
his
mic
I
have
a
question
for
you:
I
can.
So
when
do
you
think
we
can
have
an
open
source
cryo
models
in
on
GitHub.
B
I
actually
already
started
like
came
up
with
some
kind
of
like
a
thicker
thin
oxide
models
for
Angela
and
peachable
devices
for
the
standard
threshold
voltage.
So
the
I
think
we're
waiting
for
nist
to.
B
To
come
up
with
something
because
they
already
have
some
preliminary
models
and
I
think
they
are
waiting
for
some
internal
paperwork.
I
see
so
I
think
that's
the
the
main
thing
that's
like
holding
it
back
at
the
moment,
I
mean
I.
We
don't
have
the
models
for
every
single,
like
version
of
the
mosfest
so
but
for
thin
oxide
and
channel
most
best.
B
I
have
a
predominant
model
for
the
the
standard
threshold
and
height
high
threshold
voltage
devices
and
for
the
thin
oxide
feature
almost
as
I
have
a
model
for
the
standard
Treasure
World,
which
mosfet
and
like
more
recently
I'm
working
on
the
thick
oxide
devices,
but
that
having
those
models
on
like
available
to
everybody,
I
think
is
mainly
determined
by
the
paperwork.
C
C
C
My
first
remark
is
that
I
mean
that's
okay,
when
working
with
you
know
130
nanometer
technology,
but
when
you
go
down
to
very
advanced
technology,
you
simply
can't
reach
strong
immersion
or
very
strong
inversion
anymore
because
of
the
voltage
scaling
right
so
working
down
to
0.8,
Volt
or
even
below
it
becomes
actually
almost
impossible
to
to
get
into
week
to
get
into
strong
inversion.
B
C
Yeah
I
saw
that
yeah
all
right
now,
when
you
talk
about
mismatch
here
when
we
look
at
the
right
curve
on
the
figure
on
the
right,
so
those
are
really
two
devices
measured
and
and
and
you
compare
two
identical
device
or
matched
devices.
These.
C
A
C
B
C
B
I
think
one
pair
here
is
actually
is
on
the
same
die,
but
not
every
I
I
try
to
see
it's
like
I
think
there
are
like
maybe
six
colors
here
or
maybe
five,
so
the
two
of
them
are
on
the
same
die.
I
think
the
other
ones
were
in
a
different
die.
It's
been
a
long
time.
This
was
in
IBM
chip
by
the
way,
but
now
that
process
kind
of
gone.
C
So
I
mean
it's
it's
known
to
everyone
that
you
know
variations
that
particularly
here
we
see
threshold
variations.
This
is
definitely
not
good
for
for
matching
IV
characteristic.
Now,
just
want
to
say
that
there
are
cert
I
mean
designing
into
sub
threshold
requires
different
ways
to
to
make
sure
that
you
you
you
you
you
match
the
current
accurately
right,
so
you
cannot
just
design
by
imposing
the
voltage.
C
So
you
need
to
use
correct
bias
circuits
in
order
to
minimize
the
the
variation
of
the
of
the
parameters
such
as
the
threshold
voltage
right,
so
I
would
just
want
to
mitigate
a
little
bit
what
you
say.
A
C
Is
very
large
variations
and
this
is
not
improving
right
or
not,
but
it's
still
possible
to
use
sub
threshold
region
even
for
analog
circuit
design,
and
actually
we
want
to
take
advantage
of.
You
know
the
the
steeper
slope
we
want
to
take
advantage
of
the
lower
voltage
and
and
that's
the
region
which
becomes
really
interesting
when
moving
down
to
lower
temperature.
That's
where
you
save
current,
that's
where
you
can
reduce
the
voltage.
You
can
reduce
the
power,
so
we
definitely
want
to
go
there.
C
I
know
I
mean
designing
in
sub
threshold
I'm.
Talking
now
about
analog
circuits
requires
to
use
really
different
ways
to
buy
as
your
your
matched
transistors.
So
so,
in
order
to
have
you
definitely
not
bias
using
voltage
references
but
making
sure
that
your
bias
with
the
correct
current
reference
in
in
moderate
or
weak
inversion,
yeah.
C
B
C
Yeah
definitely-
and
this
is
yet
another
remark
for
analog
designers-
you
you,
if
you
don't
need
to
if
you
have
enough
speed
and
usually
when
we
move
to
very
Advanced
Technologies,
even
for
the
65,
you
usually
have
enough
speed,
don't
use
minimum
Channel
links,
definitely
yeah
for
for
self
gain
for
matching
for
all
good
reasons.
There
is
no,
if
you
have
enough
speed,
if
you
don't
go
to
RF,
don't
use
minimum
Channel
lengths,
that's
for
sure,
yeah
and
things
improve
definitely
yeah.
C
All
right,
I,
don't
know
if
I
can
ask
a
second
question:
yeah.
C
I
was
really
intrigued
about
your
noise
measurements
and
I
was
wondering
so
you
talked
a
lot
about
rtn.
You
mentioned
that
rtn
becomes
actually
quite
a
challenge,
particularly
at
low
temperature,
which
is
what
we
see
here.
B
So
the
so
rtn
is
not
obvious.
Basically,
this
kind
of
limits,
the
the
minimum
like
the
currents
like
the
depends
on
the
What
charges
they
want
to
like
like
be
able
to
measure.
So
these
limits
the
minimum
charge
they
can
sense.
B
C
C
Yeah,
did
you
have
the
opportunity
to
measure
White
Noise,
so
thermal
or
short,
noise
at
low
temperature
iron.
B
C
B
Aesthetic
connection
I
mean
the
law
our
longest
captured
waveform
was
about
like
10
15
minutes,
but
to
getting
a
10-15
minute.
Waveform
was
taking
like
hours,
sometimes.
A
Right,
yeah
thanks!
Thank
you
guys.
Is
there
any
other
questions?
I
think
we
are
over
in
an
hour
but
I'm,
okay,
taking
more
questions,
if
not,
if
not,
then
I
just
want
to
ask
Christian
if
you
would
be
happy
to
present
next
time
about
the
results,
the
measurement
results
and
what
is
the
timeline
for
that?
A
So
I
think.
Last
time
we
discussed
to
do
that
after
isscc,
but
you.
C
Know
it
depends
on
right,
so
we're
right
in
the
middle
of
doing
the
measurements
right
now
so
I'm
not
sure
we
will
have
lots
of
things
to
show.
Would
that
be
next
week
or
or
when.
A
Oh
so
that
would
be
in
two
weeks:
I
guess
so,
but
that's
very
close
to.
C
Us
we
yeah,
we
could
probably
already
show
some
preliminary
measurements.
Yes,
this
we
certainly
can
do
so
we're
right
doing
the
measurement
right
now,
so,
okay,
hopefully
by
within
two
weeks,
we
have
something
to
show,
maybe
not
all
the
extraction
part,
but
at
least
some
measurements
we
could
show.
Yes
sounds
good.
C
A
Yeah,
that's
that's
perfect.
I
think
the
14th
looks
like
a
good
date,
so
maybe
we
can
Target
that
if
not,
we
can
just
push
it
after
isscc,
okay,
cool!
Thank
you
so
much
everyone
thanks,
Christian
and
Akin.
Thank
you
excellent
presentation
and
talk
to
you
soon.
Bye,
bye,
bye,.