Latest / Talking Science / Semiconductors – The Material That Built the Modern World

Transcript

Semiconductors – The Material That Built the Modern World · Talking Science

0:00
Look at almost any device around you right now
0:02
that does something clever. You know, your phone,
0:04
your laptop, maybe the microwave in your kitchen.
0:07
Right. Or even just like the card reader at the
0:09
local coffee shop. Exactly. They all serve completely
0:11
different purposes, built with totally different
0:13
designs. But at the heart of every single one
0:15
of them is the exact same quiet miracle. A tiny
0:20
sliver of silicon etched with just unimaginably
0:24
small patterns. Welcome to this deep dive. Thanks
0:27
for having me. Today, our mission is to completely
0:29
demystify what is fairly called the Silicon Age.
0:33
And we're going to do this by slowing right down,
0:36
really unpacking the fundamental building block
0:38
of the modern world. And we're assuming you,
0:41
the listener, have absolutely zero prior science
0:44
background. Right. None at all. But by the end
0:46
of our time together, you will understand exactly
0:48
how something as basic as. Well, refined sand
0:51
ended up running basically everything. It's crazy
0:54
when you think about it. Silicon is a genuinely
0:57
strange substance. I mean, the story of how humanity
1:01
learned to exploit its specific quirks is it's
1:05
literally the story of how our modern civilization
1:06
was built. OK, so to wrap our heads around silicon.
1:11
I guess we first have to start with how different
1:13
materials in the physical world handle electricity.
1:16
Yeah, that's the best starting point. Materials
1:19
generally fall into two very distinct camps when
1:21
they're dealing with electrical current. Okay.
1:23
On one side, you have your conductors. Copper
1:26
is the classic example here, along with, you
1:28
know, most metals. Right, because their atomic
1:31
structure allows electric current to just flow
1:34
through them with virtually zero resistance.
1:36
Exactly. The wires in the walls of your house
1:39
are copper for that exact reason. They move power
1:43
from the grid to your outlets totally effortlessly.
1:47
Effortlessly, right. And then what's on the other
1:49
side? Well, on the opposite end of the spectrum,
1:51
you have insulators. Materials like rubber, glass,
1:54
plastic. Guff that blocks electricity. Yes. Their
1:57
atomic structure tightly binds electrons, so
2:00
it completely blocks electric current from flowing.
2:03
Which is why those copper wires in your walls
2:05
are wrapped in a plastic coating. Right. So the
2:08
electricity stays in the wire and doesn't just,
2:11
you know, leak out into the wooden frame of your
2:13
house and burn it down. Right. OK, let's unpack
2:15
this. We have conductors that let electricity
2:18
flow and insulators that stop it cold. But today's
2:23
deep dive is entirely focused on a semiconductor.
2:26
Yeah. A semiconductor sits right in the middle
2:29
between those two extremes. It connects electricity,
2:33
but it does a remarkably poor job of it. See,
2:36
on the surface, that sounds totally useless.
2:38
It does, right. It's like hiring an employee
2:40
who is deliberately mediocre at their one specific
2:43
job. I mean, if we want electricity to flow through
2:46
a device, why bother with a material that can't
2:49
decide what it wants to be? That's a totally
2:51
fair question. Why not just use copper? Because
2:54
that indecisive nature is actually the entire
2:56
point. What makes a semiconductor so incredibly
2:59
valuable is that its willingness to conduct electricity
3:01
can be externally controlled. Oh, interesting.
3:04
Right. With a copper wire, it's always conductive.
3:07
You can't just tell copper to suddenly stop acting
3:09
like copper. It's just always on. Exactly. But
3:12
a semiconductor is sitting on the fence. We can
3:15
push it one way or the other. And being able
3:17
to turn a material's conductivity on and off
3:19
incredibly fierce is the foundational requirement
3:21
for building a switch. A switch. Which brings
3:25
us back to silicon, I guess, the most famous
3:27
semiconductor on Earth. Yep, silicon. It is wonderfully
3:31
abundant, mostly because it's, well, the main
3:34
ingredient in ordinary sand. But wait, we can't
3:36
just scoop up sand from the beach and stick it
3:38
in a motherboard, right? Oh, definitely not.
3:40
Raw silicon, even after it has been completely
3:43
purified into a pristine, flawless crystal lattice,
3:47
is basically just a weak insulator. So it doesn't
3:50
really do much on its own? No, not at all. The
3:52
real magic requires a laboratory process. called
3:54
doping. Doping, like adding performance enhancers.
3:57
Kind of. It involves deliberately introducing
4:00
tiny amounts of other elements into that otherwise
4:02
pure silicon structure. And we are talking about
4:05
adding maybe a few atoms of an impurity for every
4:08
million or even billion atoms of silicon. Wait,
4:10
really? So we spent immense effort purifying
4:12
sand, getting it flawlessly clean just to intentionally
4:16
contaminate it. Yeah, that's exactly what we
4:19
do. That's like adding a single microscopic drop
4:21
of food coloring to an Olympic swimming tool.
4:24
That is a perfect analogy. And depending on the
4:26
specific element you use to contaminate the silicon,
4:29
you fundamentally alter its electrical personality.
4:32
How so? Well, if you add an element like phosphorus,
4:35
for instance, you create what we call N -type
4:38
silicon. Okay. N -type. The N stands for negative.
4:42
That's because phosphorus... introduces a slight
4:46
surplus of free -floating electrons into the
4:48
material. And since electrons carry a negative
4:51
charge, they are super eager to just move around.
4:54
Okay, that makes sense. Extra electrons equal
4:55
a negative charge. Right. But if you dope the
4:58
silicon with an element like boron, you create
5:01
p -type silicon. And the p stands for positive.
5:04
You guessed it. Boron actually creates a deficit
5:07
in the silicon lattice. It literally creates
5:10
empty spaces, we call them holes, where an electron
5:13
ought to be but just isn't. Oh, okay. And in
5:16
the physics of semiconductors, those empty holes
5:19
actually act as positive charges. Wait, wait,
5:21
you lost me there. If a hole is literally just
5:24
the absence of an electron, how can empty space
5:27
have a positive charge? I mean, it's literally
5:29
nothing. Yeah, that is a really great distinction
5:31
to make. It's not the empty space itself that
5:35
carries the charge. Think about the silicon atoms
5:37
surrounding that hole. Okay. They still have
5:40
their full complement of positively charged protons
5:43
sitting in their nuclei. Right. Normally, those
5:47
protons are perfectly balanced by negatively
5:49
charged electrons in orbit. But because doping
5:52
created a missing electron, that local area of
5:56
the atom now has more positive protons than negative
5:59
electrons. Oh, I see. So the net charge becomes
6:01
positive. The hole is just, you know, the easiest
6:03
way to visualize that positive vacancy moving
6:06
around as electrons jump in and out of it. I
6:08
totally get it now. The underlying structure
6:10
provides the positive charge because the balancing
6:12
electron is missing. So we have N -type silicon
6:15
packed with extra wandering electrons and P -type
6:18
silicon packed with positive vacancy. Exactly.
6:21
But sitting on a desk separately, neither chunk
6:23
does anything particularly useful. Okay. So when
6:26
does it become useful? The revolution happens
6:29
when you bond a piece of N -type silicon directly
6:32
against a piece of P -type silicon. Oh, wow.
6:34
What happens at the border where they touch?
6:37
Imagine the N -type silicon as this crowded parking
6:40
lot with too many cars frantically driving around.
6:43
Sounds stressful. It is. And the P -type silicon
6:45
is an adjacent parking lot with plenty of empty
6:48
parking spaces. The moment you push these two
6:50
lots together, the extra cars, the electrons,
6:53
immediately rush across the border to fill the
6:56
empty spaces on the other side. Huh. Wait, wouldn't
6:59
that seem like it would just neutralize the whole
7:01
thing? It neutralizes the border area, yeah.
7:03
By rushing across, those electrons lock themselves
7:07
into the holes right at the junction. Okay. And
7:10
they create this localized gridlock, a dead zone
7:13
we call the depletion region. Because all the
7:15
free electrons are now stuck in holes, there
7:17
are just no loose charge carriers left in that
7:20
middle border. Wow, okay. And that gridlock zone
7:23
acts as a physical wall. completely blocking
7:26
any further electrical current from flowing across
7:28
the silicon. So by just putting them together,
7:30
they instantly create an internal traffic jam
7:32
that stops electricity? Exactly. But we can break
7:36
that traffic jam. How do you do that? If we hook
7:39
a battery up to this piece of silicon and apply
7:41
voltage in a specific direction, the electrical
7:44
pressure actually forces the gridlocked electrons
7:48
out of their holes. Oh, so it just blows through
7:51
the wall. Yep. It instantly collapses that wall
7:53
and allows current to flow freely. But if we
7:56
flip the battery around and reverse it, the pressure
7:59
actually widens the gridlock, making the wall
8:02
even thicker. Wait, so you've basically just
8:04
created a one -way valve for electricity and
8:08
with absolutely zero moving parts. That's exactly
8:11
it. Current can push through the gridlock in
8:13
one direction but gets completely blocked in
8:15
the other. Right. And that PN junction, that
8:17
tiny one -way valve, is honestly the seed of
8:20
the entire digital age. That is wild. But a valve
8:23
that just permanently sits there blocking traffic
8:25
in one direction isn't very smart. The real leap
8:28
forward happened when engineers figured out how
8:30
to actively control that junction. How did they
8:33
do that? They added the third connection. And
8:35
that created the transistor. Okay, the transistor.
8:38
How does that third connection change the game?
8:40
Well, a standard transistor basically sandwiches
8:43
three pieces of doped silicon together. Think
8:46
of it like a pipe with a mechanical gate right
8:47
in the middle. Okay, I'm picturing it. The main
8:50
electrical current really wants to flow from
8:52
one side to the other, but the depletia regions,
8:54
that gridlock we talked about, block it. Right,
8:57
the wall is up. Exactly. The third connection
8:59
acts as a control knob. When you apply just a
9:02
tiny electrical voltage to that third connection,
9:05
it alters the electrical field inside the silicon,
9:08
temporarily dissolving the gridlock. Ah. So you
9:12
apply a little control voltage, the gridlock
9:14
vanishes, and the main current just rushes through.
9:17
The switch is on in. Bingo. And the millisecond
9:21
you remove that tiny control voltage, the gridlock
9:23
instantly reestablishes itself. And the main
9:26
current is blocked again. The switch is OFF.
9:28
Exactly. And because this entire process is just
9:31
electrons shifting positions within a solid crystal,
9:34
there's no physical switch flipping up and down.
9:36
Right. There are absolutely zero moving parts
9:39
to wear out or create friction. None at all.
9:42
A modern transistor can flip between on and off
9:45
for billions of times every single second. Billions
9:48
of times. That is it's as hard to even fathom.
9:50
And that brings us to how we get from physics
9:52
to software. Right. Right. Because most of you
9:55
listening probably know that computers use binary
9:57
software is written in ones and zeros. But there
10:00
is a massive conceptual gap between a physical
10:04
silicon switch. and like a program running on
10:08
a screen, how do we cross that bridge? Yeah,
10:11
so that bridge is constructed by wiring these
10:13
transistors together into specific patterns.
10:15
We call them logic gates. Logic gates. Right.
10:19
Imagine you have a wire carrying electricity
10:21
and you place two transistors sequentially along
10:24
that same wire. Like two toll booths on a single
10:27
lane of a highway. That is a great way to picture
10:29
it. For the electricity to make it to the end
10:31
of the wire, you have to apply a control voltage
10:33
to the first transistor to open it. And you have
10:36
to apply a control voltage to the second transistor
10:38
to open it. So if either one is closed, the current
10:40
just stops. Right. And you have just physically
10:42
built an AND gate. Oh, wow. So if I wanted an
10:45
OR gate, I would, like, split the wire into two
10:48
parallel tracks, put a transistor on each track,
10:51
and then merge them back together. You've got
10:53
it perfectly in that arrangement. Current reaches
10:55
the end if you open the first transistor or the
10:58
second transistor. That is amazing. By arranging
11:00
the physical tracks and switches, you are literally
11:02
hard -coding logical rules into the silicon.
11:05
So when you string thousands or, I guess, millions
11:09
of these AND, OR, and NOT gates together... You
11:12
create circuits that can add numbers, store information
11:15
in memory, and execute really complex sets of
11:18
instructions. So every pixel on a screen, every
11:20
saved password, every artificial intelligence
11:22
calculation... It's all just an unimaginably
11:26
vast arrangement of those microscopic logic gates
11:28
routing electrical traffic based on the rules
11:31
we built. Exactly. The entire edifice of digital
11:34
technology rests on the controllable conductivity
11:36
of that doped silicon junction. But, I mean,
11:39
making one switch or a handful of logic gates
11:42
isn't enough to build a pocket calculator, let
11:44
alone a smartphone. No, definitely not. Early
11:47
computers used really bulky individual switches.
11:51
whether they were fragile glass vacuum tubes
11:53
or early hand -wired transistors. Right, and
11:56
they took up entire rooms just to do basic math.
11:59
Exactly. The only way forward was to scale up.
12:02
Okay, so how did they do that? Well, engineers
12:04
realized they couldn't keep soldering physical
12:07
wires between individual components. It was just
12:10
too slow and too big. They needed to put all
12:13
the transistors and all the microscopic wiring
12:15
connecting them onto a single flat wafer of silicon.
12:18
Oh, and this was the birth of the integrated
12:20
circuit or the microchip. Yep. And this marks
12:23
the beginning of Moore's law, too. Right. The
12:25
famous observation by Gordon Moore that the number
12:27
of transistors we could fit onto a silicon chip
12:30
was roughly doubling every two years. Exactly.
12:33
That compounding growth was driven by this totally
12:35
relentless campaign to shrink the manufacturing
12:38
blueprint. Today, a modern microchip the size
12:41
of your fingernail holds tens of billions of
12:44
individual transistors. Okay, let's focus on
12:46
that for a moment because the scale is just baffling.
12:49
How do you actually mass produce something that
12:51
small? It's extremely difficult. I mean, a single
12:54
transistor today is far smaller than a speck
12:56
of dust. You can't use tiny mechanical tools
12:59
to carve something measured in atoms. No. You
13:01
can't. You have to use light. Light. Yeah. The
13:04
process is called photolithography, and it is
13:08
effectively atomic photography. Okay, walk me
13:10
through that. Sure. You take your pure silicon
13:12
wafer, dope it to create your P -type or N -type
13:16
foundation, and then you coat it in a light -sensitive
13:19
chemical called a photoresist. Ah, so it really
13:22
is like preparing photographic film. Precisely
13:25
like film. Yeah. You then take a photo mask,
13:27
which is basically a physical stencil of the
13:30
incredibly complex microscopic city you want
13:32
to build. Okay. You shine a specific type of
13:35
light through that stencil and project the pattern
13:38
down onto the silicon wafer. I see. The light
13:41
passes through the gaps in the stencil and hits
13:43
the photoresist chemical, altering its structure.
13:46
Then you wash the wafer in a chemical bath. which
13:49
dissolves the altered areas. Leaving your microscopic
13:52
pattern perfectly transferred onto the silicon.
13:54
Exactly. So you aren't carving the silicon directly
13:56
at all. You are projecting a map of light onto
13:59
it and using chemicals to eat away the specific
14:02
patterns. Right. And you repeat that process
14:05
dozens of times, layer by layer. That's how you
14:08
build up the complex 3D structures of billions
14:10
of transistors and the incredibly intricate copper
14:13
wiring that connects them all together. Okay,
14:15
but wait. If we are shrinking these transistors
14:18
every two years, eventually the features become
14:20
smaller than the wavelength of visible light
14:23
itself, right? Yes, that was a huge problem.
14:25
Because the stencil wouldn't cast a sharp shadow
14:27
anymore. The light would just bleed around the
14:29
edges. Exactly. And that physical limitation
14:32
forced the industry to develop extreme ultraviolet
14:36
lithography, or EUV. We basically had to abandon
14:39
visible light entirely. Wow. EUD light has an
14:42
incredibly short wavelength. which allows for
14:44
microscopic sharpness. But it presents a massive
14:48
engineering nightmare. Why is that? Because EUV
14:51
light is absorbed by absolutely everything. It
14:53
is absorbed by glass lenses, and it is even absorbed
14:56
by the air itself. Wait, what? How do you focus
14:58
a beam of light if you can't use glass lenses
15:00
and it gets absorbed by the air? You have to
15:02
build the entire machine inside a high vacuum
15:04
chamber, completely removing all the air. That
15:06
sounds insanely expensive. Oh, it is. And instead
15:10
of lenses... you use the flattest, most highly
15:13
polished mirrors humanity has ever created. These
15:16
mirrors basically bounce the EUV light back and
15:19
forth, focusing it through the stencil, onto
15:22
the silicon. That is incredible. To give you
15:24
an idea of how flat they are, if you scaled one
15:26
of these mirrors up to the size of an entire
15:28
country, the tallest bump on its surface would
15:31
be less than a millimeter high. Oh my God. Well,
15:33
this totally explains why the physical manufacturing
15:35
environment is so fiercely protected. Yeah, it
15:38
has to be. I mean, if you are using extreme ultraviolet
15:41
light to put wires that are only a few dozen
15:45
atoms wide, the slightest contamination is disastrous.
15:49
Disastrous is the right word. The factories that
15:52
house these machines, they're called fabs, have
15:54
to be significantly cleaner than a hospital operating
15:57
theater. If a single microscopic stray particle
16:00
of dust or even just a flake of dead skin from
16:04
a worker floats through the air and lands on
16:07
the silicon wafer during the printing process.
16:09
It acts like a boulder on a highway. Yes. It
16:12
casts a shadow during the lithography step, preventing
16:15
the light from hitting the photoresist. And that
16:17
shadow basically breaks the microscopic wire
16:19
you were trying to print. Exactly. And because
16:21
billions of transistors are sequentially wired
16:24
together in these complex logic gates, breaking
16:27
one critical wire can ruin the entire chip. You
16:31
know, this really clarifies a geopolitical puzzle
16:33
for me. I so. Well, we frequently hear about
16:36
global chip shortages disrupting everything from
16:39
car manufacturing to national defense. And I
16:41
always wondered, if the main ingredient is just
16:43
silicon derived from sand, why is there a bottleneck?
16:46
Why can't any country just build factory and
16:48
print their own chips? Because the bottleneck
16:50
isn't the sand. It is the sheer mind -bending
16:54
complexity of the tools required. The machines
16:57
themselves. Right. Creating an EUV lithography
17:00
machine requires coordinating hundreds of thousands
17:04
of specialized parts. For instance, you need
17:07
lasers that shoot microscopic drops of molten
17:09
ten thousands of times a second just to generate
17:13
the EUV light in the first place. That is, I
17:15
can't even wrap my head around that. So you might
17:17
have access to all the silicon in the world,
17:19
but you don't have the multi -billion dollar.
17:21
oven necessary to bake it. Exactly. The know
17:24
-how, the supply chains, and the machines capable
17:27
of doing atomic level photolithography are concentrated
17:31
in a very small number of companies and regions.
17:33
Which makes it hugely political. Absolutely.
17:35
The entire modern economy relies completely on
17:38
these tiny switches, yet very few places on Earth
17:41
can actually mass produce the most advanced versions.
17:44
Whoever controls the manufacturing of these switches
17:47
effectively controls the future of global technology.
17:50
And looking toward that future, we are facing
17:52
an entirely different kind of bottleneck, aren't
17:54
we? A physical one. Yes, we really are. Because
17:57
we have spent 60 years relentlessly shrinking
18:00
these transistors to maintain Moore's Law. But
18:04
you can't carve half an atom. The blueprint has
18:07
to stop shrinking eventually. Physical reality
18:09
is aggressively catching up with our engineering
18:12
ambitions. The components of a modern transistor
18:15
are now so incredibly thin that they are measured
18:18
in mere handfuls of atoms. Wow. And when you
18:21
get down to the atomic scale, the reliable rules
18:23
of classical physics begin to break down, and
18:26
the strange rules of quantum mechanics take over.
18:29
Wait, how does quantum mechanics interfere with
18:30
a switch? Well, remember the depletion region?
18:33
The gridlocked wall of electrons that stops the
18:36
current when the switch is off? Right. We rely
18:39
on that wall being absolutely impenetrable. But
18:42
at the atomic scale, a phenomenon called quantum
18:45
tunneling occurs. Quantum tunneling? Yeah. The
18:48
wall of the transistor is now so physically thin
18:51
that electrons sometimes just teleport right
18:53
through it. Teleport? Literally. They just appear
18:56
on the other side. But if the electrons can bypass
18:58
the wall, the switch isn't really O off F anymore.
19:00
Current is just leaking through. Exactly. The
19:03
switch becomes leaky and completely unreliable.
19:06
And if you can't trust your switches to definitively
19:09
stay OFF, your logic gates break down and your
19:12
software just stops working. That sounds like
19:14
a massive problem. We are literally running out
19:17
of atoms to build reliable physical walls. We
19:20
really are. So does this mean the party is over?
19:22
Will our devices eventually just hit a permanent
19:24
performance ceiling? Not necessarily. It represents
19:28
a maturation of the technology, not necessarily
19:30
an end to progress. Okay, so how do we get around
19:32
it? We just can't rely on simply shrinking the
19:36
flat two -dimensional blueprint anymore. Engineers
19:38
are adapting by fundamentally changing the architecture.
19:41
In what way? Instead of laying transistors out
19:43
in a flat urban sprawl, they're building vertically.
19:46
They're stacking chip layers in three dimensions
19:49
to pack way more processing power into the exact
19:52
same footprint. Oh, that makes sense. And they
19:54
are designing highly specialized logic gates
19:57
tailored for specific tasks, like artificial
20:00
intelligence, rather than just relying on general
20:03
purpose chips. Plus, they are actively researching
20:06
new exotic semiconductor materials to eventually
20:09
replace silicon altogether. So the path forward
20:12
is forcing us to be much more creative than just,
20:14
you know, shrinking the stencil. Exactly. Wow.
20:18
Let's trace this incredible journey one last
20:20
time. We started with ordinary sand. From that
20:23
sand, we extracted silicon, a completely mediocre
20:26
material that naturally conducts electricity
20:28
very poorly. Right, and then we perfectly contaminate
20:31
that silicon through doping. introducing deliberate
20:34
flaws to create n -type regions with extra electrons
20:37
and p -type regions with empty holes. And when
20:40
we push them together, those electrons and holes
20:42
crash into each other, creating a gridlock depletion
20:45
region, a one -way valve for electricity. Yep.
20:48
Then we add a third connection to that valve,
20:50
giving us a way to selectively clear the gridlock
20:53
using a control voltage. That is the transistor,
20:56
a switch with zero moving parts. Then we arrange
20:58
those tiny physical switches into overlapping
21:00
patterns on a single wire, creating logic gates
21:03
that force electricity to do math. And finally,
21:06
we scale it up. Using extreme ultraviolet light
21:09
and flawlessly flat mirrors inside pristine vacuum
21:12
chambers, we project microscopic stencils to
21:15
carve tens of billions of these logic gates onto
21:19
a sliver of silicon the size of your fingernail.
21:21
It's honestly breathtaking. That chain of physical
21:24
manipulation is what built the digital world
21:26
we inhabit today. It really is. So the next time
21:29
you casually reach into your pocket, unlock your
21:31
phone, and check a message, spare a thought for
21:34
the invisible metropolis in the palm of your
21:36
hand. Yeah. Tens of billions of microscopic switches
21:38
carved into purified sand are flipping on and
21:41
off billions of times a second in perfect lightning
21:44
-fast coordination just to display that one notification.
21:48
Unbelievable. profound to consider for the decades
21:53
ahead. What's that? Well, for over half a century,
21:56
the entire structure of our digital reality has
21:59
been built on exploiting the highly predictable
22:02
classical properties of silicon atoms. But as
22:05
we shrink our ambitions down to their absolute
22:07
physical limits, we're entering a space where
22:10
those classical rules just... Yeah. If our future
22:14
requires us to build devices that rely on the
22:17
strange teleporting reality of quantum physics.
22:20
Yeah. How fundamentally will our technology and
22:22
our relationship with it have to shift to accommodate
22:24
that new frontier? Man, that is the boundary
22:27
we are approaching next. Something to really
22:29
mull over. Thank you for joining us on this deep
22:32
dive into the Silicon Age. Keep wondering, keep
22:34
exploring, and we'll see you next time.