05 / The Horizon

The next decade will not be downloaded. It will be fabricated.

New transistor architectures, AI-driven manufacturing, geopolitical competition, and resource constraints are reshaping what it means to build chips.

Last updated June 2, 2026.

05 / The horizon

The frontier is a stack.

The next decade is being shaped by device architecture, manufacturing intelligence, geopolitics, resources, talent, materials, and light.

For sixty years, the transistor followed a familiar script: make it smaller, make it faster, make more of them. The planar transistor gave way to the FinFET, and FinFETs are giving way to three simultaneous revolutions in transistor architecture.

Gate-All-Around (GAA) Arrives in Volume

The first revolution is already in production. Gate-All-Around (GAA) nanosheet transistors replace the three-sided gate of a FinFET with a gate that wraps completely around the channel, enabling superior electrostatic control and eliminating leakage that plagued FinFETs at the 3nm node and below.

TSMC's N2 (2nm) node started volume production in 4Q25 as planned, marking its first nanosheet transistor technology. TSMC describes N2 as offering 10–15% speed improvement at the same power, or 25–30% power reduction at the same speed, with greater than 15% chip density increase over the preceding N3 node. For a foundry whose economics depend on fast yield ramps, this is the bet that pays for the next decade.

Samsung was first to announce GAA production with its 3nm MBCFET architecture, but execution and customer adoption have been harder than the announcement. The gap between announcing a node and manufacturing it profitably remains the most punishing reality in semiconductor economics.

Intel's 18A node combines RibbonFET GAA with PowerVia backside power delivery, and Intel now presents it as ready for customer projects. After years of delays and node slips, Intel has staked its foundry revival on proving that 18A can move from technical milestone to trusted manufacturing platform.

Backside Power Delivery: The Biggest Architectural Shift Since FinFET

The second revolution is arguably more profound. Backside Power Delivery Networks (BSPDN) move the power rails from the front of the wafer, where they have competed for routing space with signal wires since 1960, to the backside, using nano-scale through-silicon vias.

Intel was first to production with PowerVia on its 18A node, reporting greater than 6% performance gain at equivalent power versus front-side power networks. The improvement sounds modest until you consider it is achieved purely by rearranging where electricity enters the chip, with no change to the transistor itself.

TSMC's response is Super Power Rail (SPR), introduced with A16 as part of its 2nm-class roadmap. A16 is aimed at AI and HPC designs where dense power delivery and complex signal routing are no longer separate problems. Samsung has also described backside power as part of its future foundry roadmap.

The precise timing will vary by foundry and customer ramp, but the direction is clear: backside power is becoming a shared leading-edge roadmap item, not an Intel-only experiment.

CFET: The Next Horizon After GAA

The third revolution is still in the laboratory, but it is the consensus path forward. Complementary FETs (CFETs) stack nFET and pFET transistors vertically, eliminating the horizontal n-to-p separation that consumes precious area in every standard cell. The result: dramatically reduced standard cell height and a true leap in density.

IMEC, the Belgian research institute that has guided the industry roadmap for three decades, expects to introduce monolithic CFET (mCFET) at the A7 node. At VLSI 2024 and IEDM 2024, IMEC demonstrated key process modules including Middle Dielectric Isolation (MDI) and functional mCFET with direct backside contact to the bottom pMOS device. An intermediate technology called "Outer Wall Forksheet" is expected to extend the nanosheet roadmap to the A10 node before mCFET takes over.

The naming itself tells a story. TSMC now describes A14 as scheduled for volume production in 2028, with A13 and A12 scheduled for production in 2029. We have crossed a psychological threshold: the nanometer era is giving way to angstrom-class branding and ever tighter system-level co-optimization.