00:00:02
semiconductor manufacturing is heavily
00:00:04
dependent on deposition techniques it is
00:00:06
one of the key aspects of the
00:00:08
semiconductor manufacturing gameplay
00:00:10
Loop
00:00:11
we add thin layers modify them maybe
00:00:13
remove some of its parts and then do it
00:00:15
all over again
00:00:17
but as transistor sizes have shrank
00:00:19
those thin layers have shranked too to
00:00:21
the point where we now have to deposit
00:00:23
these new layers atom by atom
00:00:26
Atomic layer deposition as the name
00:00:28
implies it is a special variant of
00:00:30
chemical vapor deposition tuned for
00:00:32
nanometer scale in this video we're
00:00:34
going to look at this incredible
00:00:36
nanotechnology
00:00:37
but first I want to remind you about the
00:00:39
newsletter I do try to keep it up to
00:00:41
date with new things that I've been
00:00:42
hearing the full write-up for videos you
00:00:44
haven't seen before and so on the sign
00:00:46
up link is in the video description
00:00:48
below I try to put one out every week
00:00:49
maybe two alright back to the show
00:00:52
Atomic layer deposition has a
00:00:54
well-documented history it is generally
00:00:56
acknowledged I've been independently
00:00:58
invented twice
00:00:59
in the 1960s the Soviet Union produced a
00:01:03
molecular layering technique
00:01:05
this was based on a 1952 thesis by
00:01:08
Professor Valentin Alex govski at the
00:01:11
Leningrad Institute of Technology
00:01:13
he worked together with another
00:01:14
professor as I cultov throughout the
00:01:17
1960s and 1970s to develop this
00:01:19
technique
00:01:20
molecular layering is ald in all but
00:01:23
name and it might have given the Soviet
00:01:25
Union a strong advantage in
00:01:26
semiconductor Manufacturing
00:01:28
but many prominent Soviet scientists did
00:01:31
not understand molecular layering
00:01:32
furthermore they did not believe that it
00:01:34
was even possible to create a structure
00:01:36
with atomic precision
00:01:38
when alaskovsky and kolzov attempted to
00:01:41
patent the process in April 1971 the
00:01:43
Soviet patent office remarkably rejected
00:01:46
it
00:01:46
the two and their colleagues continued
00:01:48
their work in Academia but no Industrial
00:01:50
Level activities were ever initiated yet
00:01:53
another missed opportunity for the
00:01:55
Soviet Union
00:01:56
ald as we know it today was invented in
00:01:59
Finland in 1973 Dr Tomo santola a
00:02:03
professor in electrical engineering at
00:02:05
the Helsinki University of Technology
00:02:07
was contacted by Instrumentarium oi a
00:02:10
medical instrument importer
00:02:12
Dr santola was a rising star he received
00:02:15
his Doctorate at 28 the youngest to do
00:02:17
so in the electrical engineering
00:02:19
department
00:02:20
and he had already distinguished himself
00:02:22
by inventing a thin polymer film
00:02:24
humidity sensor for the company vaisala
00:02:26
the humacap The Invention is still
00:02:29
widely used today
00:02:31
Instrumentarium wanted to develop their
00:02:33
own technology products and invited
00:02:35
santola to establish a research group to
00:02:38
literally just find whatever such an
00:02:40
open-ended goal was only possible based
00:02:42
on suntola's prior success
00:02:44
eventually santola and his team proposed
00:02:47
two promising areas of study ion
00:02:49
selective sensors and electroluminescent
00:02:52
flat panel displays
00:02:53
the latter are a type of flat panel
00:02:55
display different from LEDs or oleds
00:02:58
they're quite compact power efficient
00:03:00
and deliver high contrast and brightness
00:03:03
santola's presentation to
00:03:04
instrumentariums management had been
00:03:06
very abstract the CEO was literally like
00:03:09
I am still confused but approved it
00:03:11
anyway santola started work on the
00:03:13
displays with a colleague working on the
00:03:15
sensors
00:03:16
many other people had before tackled the
00:03:19
problem of fabricating flat panel
00:03:21
displays at the time semiconductor
00:03:23
manufacturers use one of two techniques
00:03:25
for layering on these thin material
00:03:27
films sputtering and thermal evaporation
00:03:31
in the former we accelerate ions towards
00:03:34
a Target material like a metal the
00:03:36
impact ejects small particles of the
00:03:38
metal which fly out towards the desired
00:03:40
location where it accumulates as a thin
00:03:42
film in the latter we heat up a material
00:03:45
like a metal in a vacuum chamber to its
00:03:47
evaporation point
00:03:49
it vaporizes and travels over to the
00:03:51
Target location where it accumulates as
00:03:53
a thin film
00:03:55
Dr suntola knew from his previous
00:03:57
experience with thin films that these
00:03:59
tools were not sufficient you did not
00:04:01
have enough control over the thin films
00:04:03
in early June 1974 santola was pondering
00:04:06
over this while in his lab they had not
00:04:08
moved in the equipment yet so it was
00:04:10
mostly just tables and a periodic table
00:04:12
up on the wall
00:04:14
looking up on that table the idea came
00:04:16
to him
00:04:18
looking at the periodic table and
00:04:20
thinking of the overall symmetry in
00:04:21
nature to me came the idea of serving
00:04:24
the complementary elements of a compound
00:04:26
sequentially one at a time onto a
00:04:28
surface this is the Crux of atomic layer
00:04:31
deposition though back then they called
00:04:33
it Atomic layer epitaxy
00:04:36
like his big brother chemical vapor
00:04:38
deposition Atomic layer deposition
00:04:40
brings together various chemical
00:04:42
precursors
00:04:43
the big difference is in how we bring in
00:04:46
the chemicals to the substrate rather
00:04:48
than just throwing them all in there
00:04:49
like an instant pot and letting them go
00:04:51
to town we introduced the reactants one
00:04:53
after the other through a set of pulse
00:04:55
and Purge actions first we pulse a
00:04:59
primary reactant onto the substrate some
00:05:02
of this reactant gets chemically
00:05:04
absorbed into said substrate
00:05:06
then we purge the reactant by flushing
00:05:09
an inner gas through the reactor chamber
00:05:11
it cleans off any reactants that has not
00:05:14
been absorbed into the substrate this is
00:05:16
our first pulse Purge cycle
00:05:18
now we do a second round of pulse Purge
00:05:21
actions but with a different chemical
00:05:24
this one will react with the first
00:05:26
absorbed reactant to form just a portion
00:05:28
of our deposited layer usually a single
00:05:31
atom thick
00:05:32
we repeatedly run this cycle until we
00:05:34
achieve a layer of the appropriate
00:05:36
thickness
00:05:38
this is all done in a very controlled
00:05:39
environment the temperature in
00:05:41
particular is important high enough to
00:05:43
facilitate the absorption of the first
00:05:45
reactant into the substrate but not high
00:05:47
enough to cause uncontrolled reactions
00:05:50
so imagine it is kind of like pouring
00:05:52
batter into a cake pan with cvd we dump
00:05:55
all the contents into the pan that is
00:05:57
spread out over the width of the pan via
00:05:59
its own devices this method works and
00:06:02
might work faster but at the same time
00:06:04
the batter might be unevenly spread
00:06:07
across the pan especially if the pan
00:06:09
itself has some bumpiness of its own
00:06:12
but if we were to slowly spoon in the
00:06:14
batter into the pan meticulously
00:06:17
flattening it out every time we do so
00:06:19
then it is far more likely to get a
00:06:21
clean evenly distributed layer
00:06:24
the basic form of ald is to have these
00:06:27
two alternating A and B steps but more
00:06:30
advanced versions of the technique can
00:06:32
have multiple steps the additional steps
00:06:34
of this ABC ald cycle can be used to
00:06:37
modify the temperature environment or
00:06:39
the materials properties
00:06:41
you can also have ald super Cycles with
00:06:44
multiple ald Cycles to get a particular
00:06:46
thin film
00:06:48
after Dr suntola came up with the idea
00:06:50
turning it into reality came relatively
00:06:53
quickly the first experiments took place
00:06:55
in August 1974 laying down a thin layer
00:06:58
of the light emitting zinc sulfide it
00:07:00
worked right off the bat though slower
00:07:02
than expected and produced fantastic
00:07:04
display layers
00:07:06
electroluminescent displays need ald to
00:07:09
produce dense layers without any small
00:07:11
voids or bubbles called pinholes in the
00:07:13
industry such things help the layers
00:07:15
last longer in working conditions
00:07:18
his team filed the first patent in
00:07:20
November 1974 in Finland with the US
00:07:23
Japan and the Soviet Union coming
00:07:25
thereafter
00:07:26
Instrumentarium realized that this was
00:07:28
out of their wheelhouse and sold the
00:07:30
project to the TV maker loha loja was
00:07:33
able to bring more resources for
00:07:35
development the technique was not
00:07:37
unveiled to the public until 1980
00:07:39
shocking the audience with its quality
00:07:42
at first ald's only practical
00:07:44
applications were in display
00:07:46
manufacturing ald was called Atomic
00:07:49
layer epitaxy back then but the
00:07:51
semiconductor industry makers struggled
00:07:53
to accept it as an epitaxy epitaxy
00:07:56
techniques are reserved for a single
00:07:58
Crystal thin films like Silicon ald can
00:08:01
certainly do this but a struggled to
00:08:03
compete against existing epitaxy methods
00:08:05
like molecular beam epitaxy people soon
00:08:08
realize that the technique was better
00:08:10
suited for depositing non-crystalline
00:08:12
layers called amorphous layers amorphous
00:08:15
layers of dielectrics or insulators for
00:08:18
instance
00:08:19
such a thing became ald's second major
00:08:21
commercial application with the hard
00:08:23
drive industry adopted it to deposit
00:08:25
thin layers of aluminum oxide onto the
00:08:28
read write heads of their disk drives
00:08:31
considering the lack of interest in
00:08:33
epitaxy a name change was in order the
00:08:36
original first option Atomic layer
00:08:38
chemical vapor deposition or alcvd
00:08:41
didn't work out because of a copyright
00:08:43
thus they decided to go with the name
00:08:45
they have now Atomic layer deposition
00:08:48
and that has worked out very well
00:08:51
the IC makers grew interested in using
00:08:54
ald to produce very thin films capable
00:08:56
of uniformly covering substrates
00:08:58
regardless of a shape or topography
00:09:01
pioneering research in the 1990s
00:09:03
eventually led to one of the first big
00:09:05
semiconductor ald use cases producing
00:09:08
the ultra thin High K dielectric layer
00:09:11
in a planar transistor's gate
00:09:13
in a planar transistor the gate keeps
00:09:15
electrons from leaking through the gate
00:09:17
from the source to the drain
00:09:19
in 2007 Intel replaced the traditional
00:09:22
silicon dioxide with havnium oxide the
00:09:25
high K metal gate to better prevent
00:09:27
electron leakage across said gate
00:09:31
ald became the generally accepted way to
00:09:33
deposit this layer that is because they
00:09:36
have neum oxide gate layer is less than
00:09:38
five nanometers thick so we need Atomic
00:09:40
precision
00:09:42
and also with other methods like
00:09:43
sputtering Engineers worried about
00:09:45
potentially damaging the Silicon
00:09:47
substrate below the gate
00:09:49
Intel's adoption of ald largely cleared
00:09:52
the concerns that the semiconductor
00:09:54
industry previously had about the
00:09:56
technique making it an important part of
00:09:58
the manufacturing portfolio even with 3D
00:10:01
finfet Gates insulating gate oxides are
00:10:04
deposited using ald
00:10:06
and a memory ald is used to help deposit
00:10:09
thin layers on the increasingly
00:10:10
challenging topologies of modern memory
00:10:13
structures modern dynamic Ram cells have
00:10:16
a transistor and a capacitor the latter
00:10:19
carries a charge which represents either
00:10:21
a one or a zero but as those cells have
00:10:24
scaled down in size the capacitor has
00:10:27
gotten taller and skinnier some of the
00:10:29
films in these capacitors necessary for
00:10:32
the capacitor to hold its charge measure
00:10:34
has thin as six to eight nanometers
00:10:37
considering the thinness of the film and
00:10:39
the skinniness of the capacitor ald
00:10:41
works perfectly
00:10:44
in 1998 Samsung announced that they
00:10:46
would adopt ald for producing their 256
00:10:48
megabit Dynamic Ram products with their
00:10:51
180 nanometer process though it seems
00:10:54
like this was not actually adopted until
00:10:55
the one gigabit dram generation
00:10:59
IC makers have also adopted ald into
00:11:02
another significant portion of the
00:11:03
semiconductor manufacturing process back
00:11:06
end of the line processes or beol
00:11:09
these complement the front end of the
00:11:11
line or feol in front end you create the
00:11:14
transistor structures in backend you
00:11:16
connect those transistors together with
00:11:18
metal interconnects
00:11:20
in the late 1990s the semiconductor
00:11:23
industry switched from aluminum
00:11:24
interconnects to copper ones copper
00:11:27
interconnects offer Superior electrical
00:11:29
properties and reliability but also some
00:11:31
tricky manufacturing challenges one of
00:11:34
them is that copper atoms can easily
00:11:36
diffuse into the Silicon degrading it so
00:11:39
we need to First lay down a barrier
00:11:41
layer made from dielectric typically
00:11:43
made from tantalum or tantalum nitride
00:11:47
for a long time we did this with
00:11:49
sputtering or traditional cvd but as the
00:11:52
interconnects shrank we eventually
00:11:54
switched to ald to make our barrier
00:11:56
layers thin enough
00:11:57
though the old ways remained widely used
00:12:00
until at least a seven nanometer
00:12:02
generation due to concerns regarding
00:12:04
impurities left over from the chemical
00:12:06
reactions
00:12:08
this is a good opportunity to discuss
00:12:10
some of ald's downsides first as I
00:12:12
mentioned it is a technique dependent on
00:12:14
chemical reactions so you can only do
00:12:17
ald if there exists a chemical reaction
00:12:19
that you can take advantage of and as
00:12:22
with any chemical reaction you have to
00:12:24
worry about impurities this is due to
00:12:26
the Practical problems of the ald method
00:12:28
not all of the reactants absorbed into
00:12:31
the surface of the substrate end up
00:12:33
participating in the reaction good
00:12:35
processes and tools can get the impurity
00:12:37
rate down to a fraction of a percent but
00:12:40
is it within the tolerated budget that
00:12:42
is the main issue
00:12:44
second traditional ald suffers from a
00:12:46
slow rate of deposition for instance ald
00:12:49
deposits the aforementioned aluminum
00:12:52
oxide layers for a hard drive disk head
00:12:54
and about 0.11 nanometers per cycle or
00:12:58
100 to 300 nanometers per hour
00:13:01
low throughput is a serious concern for
00:13:03
any semiconductor Fab but it isn't a
00:13:06
deal breaker for a couple of reasons
00:13:08
first if you are depositing layers just
00:13:10
five nanometers thin then 100 nanometers
00:13:13
an hour ain't a huge deal
00:13:16
second you can batch these processes
00:13:18
together I remember seeing one batch
00:13:20
reactor applying an anti-tarnishing
00:13:22
layer on 2 000 jewelry pieces at one
00:13:25
time the one should note that it does
00:13:28
take longer to pulse and Purge a larger
00:13:30
chamber
00:13:32
and third we have new spatial LD tools
00:13:35
the concept has been around a while but
00:13:37
this is where you separate the pulse and
00:13:39
Purge steps by space rather than time
00:13:42
one way to implement this concept is to
00:13:45
have multiple Chambers each for handling
00:13:47
one of the pulse and perch steps with
00:13:49
the Wafers moving through them the
00:13:51
Olympia tool from Applied Materials uses
00:13:53
this concept with the Wafers on a drum
00:13:56
rotating through various Chambers
00:13:59
another big development is plasma
00:14:01
assisted ald
00:14:02
it uses plasma has the reactant in the
00:14:05
second of the two pulse and Purge Cycles
00:14:08
first encountered in the 1990s by
00:14:11
members of Philips research Labs plasma
00:14:13
assisted ald can produce better quality
00:14:16
films and also work in relatively low
00:14:18
temperatures this new technique also
00:14:21
unlocked one of ald's largest markets
00:14:23
multi-patterning I discussed
00:14:25
self-aligned double patterning or sadp
00:14:28
in a prior video
00:14:30
you run a photo lithography step
00:14:33
then you add oxide spacers silicon
00:14:36
dioxide or some other oxide directly
00:14:38
onto the lithography pattern then you do
00:14:41
your etching and remove the spacers now
00:14:44
you have lines twice as dense after just
00:14:47
a single photolithography step
00:14:50
self-aligned quadruple patterning is
00:14:52
sadp done twice
00:14:54
plasma assisted ald was ideal for this
00:14:57
because you had to be able to deposit a
00:15:00
uniformly thin layer that closely hugged
00:15:02
the patterns turns and corners
00:15:04
and since it worked at low temperatures
00:15:06
below 100 degrees Celsius it posed less
00:15:09
deformation risk for the thin layers
00:15:13
Intel first introduced sadp into the 22
00:15:16
nanometer nodes due to euv being
00:15:18
unavailable at the time the memory
00:15:21
makers followed on at the 30 to 35
00:15:23
nanometer node levels
00:15:25
even now that UV is working
00:15:27
multi-pattening isn't going anywhere
00:15:29
IMAX three nanometer node still makes
00:15:32
extensive use of 193i lithography with
00:15:35
saqp due to euv's costs and inherent
00:15:38
challenges
00:15:39
this is good for Bitcoin I mean plasma
00:15:42
assisted ald
00:15:45
Leading Edge semiconductors are getting
00:15:47
both smaller and taller for instance the
00:15:50
major foundries are moving to using gate
00:15:52
all-around transistors which are made up
00:15:54
of stacked Nano sheets totally
00:15:56
surrounded by the Silicon gate on the
00:15:59
front end of the line these structures
00:16:00
have complex topologies the challenges
00:16:03
of building and coding these is an
00:16:05
opportunity for ald to take share from
00:16:07
traditional deposition processes like
00:16:10
PVD and cvd
00:16:12
on the back end of the line we have new
00:16:14
interconnect structures like through
00:16:16
silicon Vias or tsvs basically channels
00:16:19
connecting different levels of 3D
00:16:21
integrated chips ald already plays a big
00:16:24
role in 2D copper interconnects and will
00:16:27
continue that when they go to 3D
00:16:30
the work started by Dr suntola's team in
00:16:33
the 1970s continues to mature and spread
00:16:35
throughout the industry over 50 years
00:16:37
later with Leading Edge semiconductors
00:16:39
we're moving around individual molecules
00:16:41
and atoms now ald is one of the few
00:16:44
tools capable of letting us do it all
00:16:47
right everyone that's it for tonight
00:16:48
thanks for watching subscribe to the
00:16:49
Channel Sign up for the newsletter and
00:16:51
I'll see you guys next time
