Samsung's 3nm GAA transistors and why you

Samsung just started producing chips using Gate-All-Around (GAA) transistors. This is a sentence that means almost nothing to most people and almost everything to the semiconductor industry.

I spent the weekend reading about it. I’m going to try to explain why it matters.

The shape of a transistor

A transistor is a tiny switch. On or off. One or zero. Every computation your phone, laptop, or car makes is built from billions of these switches flipping billions of times per second.

For the last 20 years, the shape of choice has been the FinFET. “Fin” because the channel (the part where current flows) sticks up vertically like a fin, and the gate wraps around three sides of it. The gate controls whether the switch is on or off. More contact between gate and channel means better control.

FinFET was introduced at the 22nm node by Intel in 2012. Before that, transistors were flat (planar). Making them 3D, adding that fin, gave the gate more surface area to work with. More control. Less current leakage. It was a breakthrough that enabled a decade of continued miniaturization.

But FinFET has limits. As you shrink the fin narrower and narrower, the channel gets so thin that quantum effects start causing problems. Current leaks in ways you can’t control. The gate’s grip on the channel weakens. You’re trying to hold water with your fingers closing tighter around a thinner and thinner stream.

At some point, the fin isn’t enough.

Enter GAA

Gate-All-Around does exactly what the name suggests. Instead of the gate wrapping around three sides of a fin, it wraps around all four sides of a flat nanosheet. Complete enclosure.

The current flows through the nanosheet. The gate surrounds it completely. No exposed surface for current to leak from. Maximum control.

Samsung’s implementation uses stacked nanosheets. Multiple thin sheets layered vertically, each wrapped by the gate. More sheets means more current capacity without making the footprint larger.

The physics are genuinely elegant. By switching from a fin (3-sided contact) to a nanosheet (4-sided contact), you get better electrostatic control at smaller dimensions. It’s the kind of solution that seems obvious in retrospect. Of course you’d want to surround the channel completely. But making it work at 3nm scale, manufacturing billions of these structures on a single wafer, reliably, at volume, is something else entirely.

Why it matters

Here’s the thing most people don’t realize about semiconductor manufacturing: the physics gets harder at every step. Each new node isn’t just “smaller.” It’s a new set of problems in quantum mechanics, materials science, photochemistry, and thermal management.

TSMC is still using FinFET at 3nm. Their N3 process squeezes more performance out of the existing architecture. Samsung decided to make the architectural jump now.

There are tradeoffs. Samsung’s initial 3nm GAA yields (the percentage of working chips per wafer) are reportedly lower than ideal. First-generation anything tends to be rough. But the ceiling for GAA is higher than the ceiling for FinFET. As yields improve, GAA should provide better performance per watt than any FinFET design can achieve.

IEEE Spectrum and AnandTech have been covering the technical details. The short version: GAA nanosheets are expected to carry the industry from 3nm to potentially 1nm and beyond.

The bigger picture

Every AI model you use, every self-driving car, every smartphone, every satellite, runs on chips made by pushing the laws of physics closer to their limits. FinFET got us from 2012 to now. GAA will get us from now to wherever “now” is in 2030.

The people who design these transistor architectures are solving problems at a scale most of us can’t visualize. We’re talking about structures measured in atoms. A nanosheet is maybe 5nm thick. That’s roughly 25 silicon atoms stacked vertically. The gate oxide layer between the gate and the channel might be 12 atoms thick.

Twelve atoms separating the switch mechanism from the current channel. And it has to work billions of times, on billions of transistors, without a single one failing for years of continuous operation.

I find this more impressive than most things humans do. Not because it’s flashy. Because it’s invisible. The most profound engineering of our era happens at a scale we literally cannot see.

Samsung just took the first step into the next era of that invisible engineering. The transition will take years. TSMC will follow with their own GAA implementation. Intel has their version (RibbonFET) planned for the 20A node.

But Samsung went first. And after spending a weekend understanding what GAA actually is, I wanted to note that. Because this is the kind of thing that changes everything and almost nobody notices.

Related thinking: