EP06 - Transistor Evolution

How transistors evolved from 2D planar to 3D FinFET to RibbonFET. Learn about short-channel effects, quantum leakage, classical scaling limits, and the architectural breakthroughs that saved Moore's Law.

Tweet coming soon

Two nanometers... that's like, ten atoms wide. But why stop there?

If smaller is better, let's just make it atom-sized. More transistors, more power. Easy.

Whoa. What was that? Did my simulation just have a seizure?

It wasn't a seizure. It was a **quantum tantrum**. Your simulation just slammed headfirst into a physical wall that the entire semiconductor industry spent a decade and billions of dollars trying to overcome.

Kurumi! I just told it to make a smaller switch. What's the big deal?

You're thinking of a transistor as a simple light switch. But it's more like a dam controlling a river of electrons. And you just tried to build a dam that's shorter than the river is deep.

This is a **Planar Transistor**. The workhorse of the industry for 40 years. The 'Source' and 'Drain' are the start and end of the river. The 'Channel' is the riverbed. The 'Gate' is a dam that presses down from above to stop the flow.

Simple enough. Gate up, flow is on. Gate down, flow is off.

Exactly. And for decades, Moore's Law was easy. We just scaled everything down. Smaller river, smaller dam. This was called **Classical Scaling**. But as we did, the 'river' got incredibly short.

The dam couldn't stop the flow anymore. The electrons just tunneled under the gate. This is called the **Short-Channel Effect**. The switch wouldn't turn off. It was always leaking.

So the chips were wasting power?

Wasting catastrophic amounts of power. They were getting so hot they risked melting. By the mid-90s, everyone could see the end of the road. The planar transistor had a hard physical limit. Moore's Law was about to die.

The industry was in a panic. DARPA, the US military's research agency, put out a desperate call in 1996: 'Find us a 25-nanometer switch that actually works.' They knew the entire future of computing was at stake.

So who solved it?

A team at UC Berkeley, led by Professor Chenming Hu. He had seen this problem coming for years. He proposed two ideas. The second one, an idea he sketched on a long flight, changed everything.

Professor Hu realized the problem was control. The gate only touched the river from the top. It was like trying to stop a river with a single paddle. His idea was simple and brilliant: 'What if we turn the river on its side?'

This is the **FinFET**. Fin-Field-Effect-Transistor. Instead of a riverbed, the electrons now flow through a tall, thin wall of silicon.

Because now, the dam can wrap around it.

The gate now has three times the surface area, and therefore three times the control over the channel. It's not just paddling from above; it's squeezing the flow from all sides.

This one architectural change saved the semiconductor industry. Intel was the first to mass-produce it in 2011, calling it their 'Tri-Gate' transistor. It allowed them to keep shrinking chips for another fifteen years.

It was. But the industry is relentless. They kept making the fins thinner and thinner to pack them closer together.

Eventually, the fin became so thin—just a few atoms across—that we started losing control again. The gate couldn't get a good 'grip'. The quantum leaks started coming back.

So... we were back at the same wall?

The exact same wall. We needed even more control. Squeezing from three sides wasn't enough. We needed to squeeze from *every* side. This led to the next great architectural leap, pioneered by companies like Samsung.

The next idea was even more radical. What if we laid the fin down, sliced it into several horizontal layers, and floated them on top of each other?

This is a **Gate-All-Around FET**, or what Intel calls a **RibbonFET**.

So it's like a multi-level highway overpass?

Exactly. The gate now envelops the channel from every possible direction. It is the ultimate grip. The leaks stop. This is the technology inside that '2-nanometer' chip.

So... Moore's Law isn't just about shrinking. It's about shape. It went from a 2D floor, to a 3D wall, to a stack of 3D ribbons. We're building quantum skyscrapers.

Precisely. And with each new layer of architectural complexity, the cost and difficulty of manufacturing explodes.

Your simulator crashed because of physics. An entire company can crash because of economics. The biggest challenge isn't designing these things. It's **Yield**.

This wafer is your city. Each chip might sell for $1000. But the process is so sensitive that a single atomic flaw can kill a chip.

At these scales, the failure rate is astronomical. A 'good' yield might be 80%. A bad one can be 10%. Can you imagine a factory where nine out of every ten things you make go straight into the garbage?

So my simulation didn't crash because 0.1nm is impossible... it crashed because the cost of failure approaches infinity.

It crashed because it didn't have to justify its own existence to a board of investors. The real world is a brutal battle between the laws of physics and the laws of economics.