- Hi, I'm Carrie Anne, and welcome to "CrashCourse: Computer Science."
Over the past six episodes, we've delved into software, from early programming efforts to modern software engineering practices.
Within about 50 years, software grew in complexity from machine code punched by hand onto paper tape to object-oriented programming languages compiled in integrated development environments.
But this growth in sophistication would not have been possible without improvements in hardware.
(indie rock music) (chiptune music) To appreciate computing hardware's explosive growth in power and sophistication, we need to go back to the birth of electronic computing.
From roughly the 1940s through the mid-1960s, every computer was built from individual parts called discrete components which were all wired together.
For example, the ENIAC consisted of more than 17,000 vacuum tubes, 70,000 resistors, 10,000 capacitors, and 7,000 diodes, all of which required 5 million hand-soldered connections.
Adding more components to increase performance meant more connections, more wires, and just more complexity, what was dubbed the Tyranny of Numbers.
By the mid 1950s, transistors were becoming commercially available and being incorporated into computers.
These were much smaller, faster, and more reliable than vacuum tubes, but each transistor was still one discrete component.
In 1959, IBM upgraded their vacuum-tube-based 709 computers to transistors by replacing all the discrete vacuum tubes with discrete transistors.
The new machine, the IBM 7090, was six times faster and half the cost.
These transistorized computers marked the second generation of electronic computing.
However, although faster and smaller, discrete transistors didn't solve the Tyranny of Numbers.
It was getting unwieldy to design, let alone physically manufacture computers with hundreds of thousands of individual components.
By the the 1960s, this was reaching a breaking point.
The insides of computers were often just huge tangles of wires.
Just look at what the inside of this PDP-8 from 1965 looked like.
The answer was to bump up a new level of abstraction, and package up underlying complexity.
The breakthrough came in 1958 when Jack Kilby, working at Texas Instruments, demonstrated such an electronic part, "wherein all the components of the electronic circuit "are completely integrated."
Put simply, instead of building computer parts out of many discrete components and wiring them all together, you put many components together inside of a new single component.
These are called integrated circuits, or ICs.
A few months later in 1959, Fairchild Semiconductor, lead by Robert Noyce, made ICs practical.
While Kilby built his ICs out of germanium, a rare and unstable material, Fairchild used the abundant silicon which makes up about a quarter of the earth's crust.
It's also more stable, therefore more reliable.
For this reason, Noyce is widely regarded as the father of modern ICs, ushering in the electronics era, and also Silicon Valley, where Fairchild was based and where many other semiconductor companies would soon pop up.
In the early days, an IC might only contain a simple circuit with just a few transistors like this early Westinghouse example.
But even this allowed simple circuits, like the logic gates from episode three, to be packaged up into a single component.
ICs are sort of like Lego for computer engineers, building blocks that can be arranged into an infinite array of possible designs.
However, they still have to be wired together at some point to create even bigger and more complex circuits like a whole computer.
For this, engineers had another innovation, printed circuit boards, or PCBs.
Instead of soldering and bundling up bazillions of wires, PCBs, which could be mass manufactured, have all the metal wires etched right into them to connect components together.
By using PCBs and ICs together, one could achieve exactly the same functional circuit as that made from discrete components, but with far fewer individual components and tangled wires.
Plus, it's smaller, cheaper, and more reliable.
Triple win.
Many early ICs were manufactured using teeny tiny discrete components packaged up as a single unit like this IBM example from 1964.
However, even when using really really itty-bitty components, it was hard to get much more than around five transistors onto a single IC.
To achieve more complex designs, a radically different fabrication process was needed that changed everything, photolithography.
In short, it's a way to use light to transfer complex patterns to a material like a semiconductor.
It only has a few basic operations, but these can be used to create incredibly complex circuits.
Let's walk through a simple, although extensive example, to make one of these.
We start with a slice of silicon, which, like a thin cookie, is called a wafer.
Delicious!
Silicon, as we discussed briefly in episode two, is special because it's a semiconductor, that is, a material that can sometimes conduct electricity and other times does not.
We can control where and when this happens, making silicon the perfect raw material for making transistors.
We can also use a wafer as a base to lay down complex metal circuits, so everything is integrated, perfect for integrated circuits.
The next step is to add a thin oxide layer on top of the silicon, which acts as a protective coating.
Then we apply a special chemical called a photoresist.
When exposed to light, the chemical changes and becomes soluble so it can be washed away with a different special chemical.
Photoresists aren't very powerful by themselves, but are super powerful when used in conjunction with a photomask.
This is just like a piece of photographic film, but instead of a photo of a hamster eating a tiny burrito, it contains a pattern to be transferred onto the wafer.
We do this by putting a photomask over the wafer and turning on a powerful light.
Where the mask blocks the light, the photoresist is unchanged.
But where the light does hit the photoresist, it changes chemically which lets us wash away only the photoresist that was exposed to light selectively revealing areas of our oxide layer.
Now by using another special chemical, often an acid, we can remove any exposed oxide and etch a little hole the entire way down to the raw silicon.
Note that the oxide layer under the photoresist is protected.
To clean up, we use yet another special chemical that washes away any remaining photoresist.
Yep, there are a lot of special chemicals in photolithography, each with a very specific function.
So now we can see the silicon again, we want to modify only the exposed areas to better conduct electricity.
To do that, we need to change it chemically through a process called doping.
I'm not even going to make a joke.
Let's move on.
Most often this is done with a high temperature gas, something like phosphorus, which penetrates into the exposed area of silicon.
This alters its electrical properties.
We're not going to wade into the physics and chemistry of semiconductors, but if you're interested, there's a link in the description to an excellent video by our friend Derek Muller from Veritasium.
But we still need a few more rounds of photolithography to build a transistor.
The process essentially starts again, first by building up a fresh oxide layer which we coat in photoresist.
Now we use a photomask with a new and different pattern allowing us to open a small window above the doped area.
Once again, we wash away remaining photoresist.
Now we dope and avoid telling a hilarious joke, again, but with a different gas that converts part of the silicon into yet a different form.
Timing is super important in photolithography in order to control things like doping diffusion and etch depth.
In this case, we only want to dope a little region nested inside the other.
Now we have all the pieces we need to create our transistor.
The final step is to make channels in the oxide layer so that we can run little metal wires to different parts of our transistor.
Once more, we apply a photoresist and use a new photomask to etch little channels.
Now we use a new process called metalization that allows us to deposit a thin layer of metal like aluminum or copper.
But we don't want to cover everything in metal.
We want to etch a very specific circuit design.
So very similar to before, we apply a photoresist, use a photomask, dissolve the exposed resist, and use a chemical to remove any exposed metal.
Whew, our transistor is finally complete.
It has three little wires that connect to three different parts of the silicon, each doped a particular way to create, in this example, what's called a bipolar junction transistor.
Here's the actual patent from 1962, an invention that changed our world forever.
Using similar steps, photolithography can create other useful electronic elements, like resistors and capacitors, all on a single piece of silicon, plus all the wires needed to hook them up into circuits.
Goodbye discrete components.
In our example, we made one transistor, but in the real world, photomasks lay down millions of little details all at once.
Here is what an IC might look like from above with wires crisscrossing above and below each other interconnecting all the individual elements together into complex circuits.
Although we could create a photomask for an entire wafer, we can take advantage of the fact that light can be focused and projected to any size we want.
In the same way that a film can be projected to fill an entire movie screen, we can focus a photomask onto a very small patch of silicon creating incredibly fine details.
A single silicon wafer is generally used to create dozens of ICs.
Then once you've got a whole wafer full, you cut them up and package them into microchips, those little black rectangles you see in electronics all the time.
Just remember, the heart of each of those chips is one of these small pieces of silicon.
As photolithography techniques improved, the size of transistors shrunk allowing for greater densities.
At the start of the 1960s, an IC rarely contained more than 5 transistors.
They just couldn't possibly fit.
But by the mid 1960s, we were starting to see ICs with over 100 transistors on the market.
In 1965, Gordon Moore could see the trend that approximately every two years, thanks to advances in materials and manufacturing, you could fit twice the number of transistors into the same amount of space.
This is called Moore's Law.
The term is a bit of a misnomer though.
It's not really a law at all, more of a trend.
But it's a good one.
IC prices also fell dramatically from an average of $50 in 1962 to around $2 in 1968.
Today, you can buy ICs for cents.
Smaller transistors and higher densities had other benefits too.
The smaller the transistor, the less charge you have to move around allowing it to switch states faster and consume less power.
Plus more compact circuits meant less delay in signals resulting in faster clock speeds.
In 1968, Robert Noyce and Gordon Moore teamed up and founded a new company, combining the words integrated and electronics, Intel, the largest chip maker today.
The Intel 4004 CPU, from episodes seven and eight, was a major milestone.
Released in 1971, it was the first processor that shipped as an IC, what's called a microprocessor, because it was so beautifully small.
It contained 2,300 transistors.
People marveled at the level of integration, an entire CPU in one chip, which just two decades earlier, would have filled an entire room using discrete components.
This era of integrated circuits, especially microprocessors, ushered in the third generation of computing.
And the Intel 4004 was just the start.
CPU transistor count exploded.
By 1980, CPUs contained 30,000 transistors.
By 1990, CPUs breached the 1 million transistor count.
By 2000, 30 million transistors, and by 2010, one billion transistors in one IC, OMG!
(laughs) To achieve this density, the finest resolution possible with photolithography has improved from roughly 10 thousand nanometers, that's about 1/10th the thickness of a human hair, to around 14 nanometers today.
That's over 400 times smaller than a red blood cell.
And of course, CPUs weren't the only components to benefit.
Most electronics advanced essentially exponentially, RAM, graphics cards, solid state hard drives, camera sensors, you name it.
Today's processors, like the A10 CPU inside of an iPhone 7, contains a mind melting 3.3 BILLION transistors in an IC roughly one centimeter by one centimeter.
That's smaller than a postage stamp.
And modern engineers aren't laying out these designs by hand one transistor at a time.
It's not humanly possible.
Starting in the 1970s, very-large-scale integration, or VLSI software, has been used to automatically generate chip designs instead.
Using techniques like logic synthesis, where whole high-level components can be laid down like a memory cache, the software generates circuits in the most efficient way possible.
Many consider this to be the start of the fourth generation computers.
Unfortunately, experts have been predicting the end of Moore's Law for decades, and we might finally be getting close to it.
There are two significant issues holding us back from further miniaturization.
First, we're bumping into limits on how fine we can make features on a photomask and it's resultant wafer due to the wavelengths of light used in photolithography.
In response, scientists have been developing light sources with smaller and smaller wavelengths that can project smaller and smaller features.
The second issue is that when transistors get really really small where electrodes might be separated by only a few dozen atoms, electrons can jump the gap, a phenomenon called quantum tunneling.
If transistors leak current, they don't make very good switches.
Nonetheless, scientists and engineers are hard at work figuring out ways around these problems.
Transistors as small as one nanometer have been demonstrated in research labs.
Whether this will ever be commercially feasible remains masked in mystery.
But maybe we'll be able to resolve it in the future.
I'm dying to know.
See you next week.
