返回首页
原创
AI 趋势
2026/06/27

Microscopic Skyscrapers: Inside IBM's 3D Chip

If you look closely at your fingernail, it’s hard to imagine fitting 100 billion microscopic switches onto its surface. Yet, that is exactly the scale of IBM’s...

Microscopic Skyscrapers: Inside IBM's 3D Chip
芯片技术
摩尔定律
IBM
硬件创新
算力

If you look closely at your fingernail, it’s hard to imagine fitting 100 billion microscopic switches onto its surface. Yet, that is exactly the scale of IBM’s latest chip prototype—a breakthrough that suggests the death of Moore’s Law may have been greatly exaggerated.

For decades, the semiconductor industry has operated on a simple mandate: shrink transistors to make computers faster. But as these components approach the size of mere dozens of nanometers, quantum mechanics begins to wreak havoc on their functionality. You simply cannot shrink them much further. To keep progress alive, IBM has stopped trying to build smaller and started building taller.

The new approach, dubbed "nanostacking," relies on a vertically integrated architecture known as a complementary field-effect transistor (CFET). Instead of laying out transistors side-by-side on a flat silicon plain, engineers are constructing microscopic two-story buildings.

While other companies have experimented with stacking by fabricating two separate chips and bonding them together, IBM’s method is more like baking a layer cake from scratch. They build the first layer of transistors, lay down more silicon, and construct the second layer directly on top. Crucially, the top layer doesn't sit perfectly flush with the bottom; it is staggered. This offset design elegantly simplifies the labyrinth of wiring required to connect everything.

However, building microscopic skyscrapers comes with a massive hurdle known as the "thermal budget." To construct the second floor without melting the delicate electrical connections of the first floor, the entire manufacturing process must be kept below 400 degrees Celsius. IBM managed to achieve this delicate temperature balance, though the exact recipe remains a closely guarded corporate secret.

The implications of this vertical leap are profound. Compared to the company's previous state-of-the-art designs from 2021, the nanostack architecture can either boost processing performance by 50% or cut energy consumption by a staggering 70%.

As artificial intelligence models grow increasingly complex, the data centers required to run them are consuming alarming amounts of electricity. A chip architecture that can drastically slash power usage while maintaining top-tier performance is exactly what the AI industry needs to sustain its current boom.

Industry analysts note that while IBM's marketing label of "0.7 nanometer" is largely symbolic—the actual distance between transistors remains around 40 nanometers—the technological leap is very real. By successfully moving into the third dimension, engineers have effectively bought the computing world another decade of exponential growth. The future of hardware, it seems, is looking up.

Key Points

  • IBM's new prototype fits 100 billion transistors onto a fingernail-sized chip by stacking them vertically.
  • The CFET architecture staggers the top layer of transistors over the bottom layer to simplify wiring.
  • Engineers had to keep manufacturing temperatures below 400°C to avoid melting the bottom layer's connections.
  • The design promises up to 50% more performance or 70% greater energy efficiency, vital for modern data centers.

Why It Matters

As AI development accelerates, it demands unprecedented computing power and electricity. By shifting to 3D chip architecture, the tech industry can continue to scale AI capabilities without hitting an immediate physical or energy wall.


Sources: