Megan Skiendel: For a decade, the semiconductor industry has been caught in a paradox: transistors keep getting smaller and denser, yet the exponential cost and power benefits that made Moore's Law meaningful have just — evaporated, replaced by billion-dollar architectural gymnastics that only like three companies on earth can actually afford.
Zara Reyes: No but wait — that's the part that gets me, because the headline is still 'Moore's Law is alive,' TSMC is still hitting new nodes, and yet the thing that made the law matter — cheaper, faster, lower power, every two years, basically on schedule — that's just gone.
Megan Skiendel: Right, and honestly I watched this exact reframe happen in real time — when I was at Intel the conversation shifted, quietly, from 'how do we shrink' to 'how do we stack,' and the people in the room knew what that meant even if the press releases didn't say it.
Zara Reyes: It's like — okay, the underlying physics literally stopped cooperating, electrons are tunneling through barriers that are supposed to stop them, and the industry's answer is 'cool, we'll just go three-dimensional and spend ten billion dollars a fab.'
Megan Skiendel: Which is a solution, technically, but it's a solution that consolidates the entire industry down to maybe two or three fabs globally — and that is a structural shift, not a blip.
Zara Reyes: So today we're getting into what actually broke, why quantum mechanics is lowkey the villain of chip scaling, and what it means that the only path forward costs more than most countries' GDP — let's go.
Zara Reyes: Transistor density is still going up. Cost per transistor is not. Those two things have been decoupled for about a decade and the industry is acting like that's fine.
Megan Skiendel: Right. And that decoupling is the whole story. Gordon Moore's original 1965 observation in Electronics Magazine — the thing that became scripture — it was always two promises bundled together. More transistors AND falling cost. The industry kept the first half and quietly dropped the second.
Zara Reyes: And nobody named it. Like, TSMC is shipping 3nm nodes, Intel is talking 1.4nm by 2026, the transistor counts keep climbing — but the cost curve broke somewhere around when we hit 7nm, 5nm, single-digit territory. That's when quantum tunneling stopped being a theoretical footnote.
Megan Skiendel: Okay so explain tunneling for a second, because this is where people lose the thread.
Zara Reyes: So in a MOSFET — a metal-oxide-semiconductor field-effect transistor, the dominant transistor design — you have a gate oxide layer that's supposed to block electrons. Keep them where you want them. But at these scales, the barrier is so thin that electrons just... go through it anyway. Probabilistically. That's gate leakage current. You're burning power on electrons you didn't ask for.
Megan Skiendel: And then there's source-to-drain tunneling on top of that — where electrons skip the gate entirely, just tunnel straight through the channel. At channel lengths below 5nm, the gate basically loses control. Classical switching breaks down.
Zara Reyes: Which is what J.R. Powell was saying back in 2008 in his paper on the quantum limit to Moore's Law. This wasn't a surprise. The physics was visible from a long way out.
Megan Skiendel: Look, here's what I'd push back on though. The industry has had 'end of Moore's Law' scares before. Every decade someone publishes the obituary. And every decade Intel or someone figures out a new trick — FinFETs, strained silicon, high-k dielectrics. So I'm a little skeptical when people say this time is genuinely different.
Zara Reyes: No but — the tricks are getting more expensive each time. That's what's different. Gate-all-around transistors, which wrap the gate around all sides of the channel to suppress leakage — that works. It actually works. But the manufacturing complexity is insane. You're not optimizing. You're rebuilding the architecture from scratch.
Megan Skiendel: And the cost doesn't distribute the way it used to.
Zara Reyes: Exactly. When Moore revised his prediction in 1975 to a two-year doubling cadence, the whole premise was that falling cost per transistor would keep the gains broadly accessible. That was the democratizing promise. 3D stacking, chiplet architecture, heterogeneous integration — those are the responses now. And they work for TSMC. They work for AMD. They do not work for companies that can't spend tens of billions on a new fab.
Megan Skiendel: Oh, honestly — Intel and TSMC hit the same tunneling wall at roughly the same time. Late 2010s. Same physics problem. Intel's market cap is roughly half what it was at its peak. TSMC's has doubled. Same fundamental constraint, completely different outcomes. That's not a physics story.
Zara Reyes: Wait — that's actually the point I keep wanting to land. The quantum tunneling problem is real. But at sub-2nm, the research shows process variation is now just as limiting. Manufacturing imprecision at the atomic scale. You're trying to place features that are tens of atoms wide and they don't land the same way twice.
Megan Skiendel: So the binding constraint shifted from physics to manufacturing yield. Which is a capex problem. Which means whoever has the deepest pockets sets the roadmap.
Zara Reyes: And that's where I think the 'Moore's Law continues in spirit' argument gets shaky. Like yes, transistor counts are still climbing. But the self-heating effects in something like a 5nm silicon nanowire transistor, the electron-phonon coupling issues — these aren't incidental. They compound. You solve tunneling with gate-all-around and you inherit a new thermal problem.
Megan Skiendel: Here's what I keep coming back to though — Moore's Law was always partly a coordination mechanism. The industry organized roadmaps around it. Suppliers, fabs, designers, everyone committed to the same timeline. The 'law' was self-fulfilling because enough powerful people agreed to make it true. So when we say it's decelerating — is the physics actually the limit, or did the coordination break?
Zara Reyes: Both. But the coordination broke because the economics of the next node stopped penciling out for most players. Mark Lundstrom at Purdue has been writing about MOSFET limits for years — the ballistic injection velocity ceiling, where you literally cannot move electrons through the channel faster than quantum mechanics allows. That's a real floor. But below that floor, it's all economics.
Megan Skiendel: Man, okay. Here's the thing I keep landing on — every time the semiconductor industry has hit a wall, the story we told ourselves was 'physics is the enemy.' And it was never really physics. It was always the economics wearing physics as a costume. The tunneling is real. The ROI problem is realer.
Zara Reyes: Right, exactly — and like, TSMC and Intel and Samsung are basically the only ones left who can even afford to find out. So the rest of the industry is already living in the post-Moore's-Law world and just hasn't named it yet. Which is honestly the more interesting story — not when does it end, but who already moved on and didn't tell anyone.
Megan Skiendel: And AI walks into all of that hungry. So — yeah. We don't know how that resolves. Nobody does. Catch you next week.
