Atoms, angstroms and AI
The future of advanced logic
Glen Wilk, Vice President, Technology, discusses why the next generation of AI will depend on controlling matter at the atomic scale.
Artificial intelligence is reshaping the semiconductor industry at an unprecedented rate. Demand for compute continues to accelerate as AI moves into every aspect of modern life, from cloud infrastructure and data centers to the devices we use every day.
This AI hypercycle creates simultaneous pressure for higher performance, greater efficiency, and faster innovation. As a result, the semiconductor industry is being reshaped by two forces: the explosive growth of intelligence usage and the need to deliver this capability within increasingly challenging power and energy constraints.
A new generation of advanced logic devices
The 2025 year marked the production ramp for gate-all-around (GAA) device technology, establishing this architecture as the foundation for future logic generations. Unlike previous transistor architectures, the GAA geometry completely surrounds the channel with the gate stack, which achieves superior electrostatic control, thereby enabling continued scaling.
Although the GAA architecture is a significant milestone for the industry, and will be in production for several technology nodes, it will evolve into yet another variant. Industry roadmaps already point toward complementary field-effect transistor (CFET) architectures, where N-type and P-type devices are vertically stacked to drive even greater density and performance. As logic moves deeper into three-dimensional architectures, complexity rises sharply, creating new demands on the materials that enable these structures and the precision with which they must be engineered.
At advanced nodes, performance variability budgets are shrinking rapidly. The shift from 2 nm technologies to Angstrom-class nodes reflects a fundamental change in how logic scales. Scaling is increasingly determined not primarily by lithography, but by how precisely materials can be engineered and controlled. The industry's shift into the Angstrom era is ultimately a move toward atomic precision, where even the smallest variations can influence device behavior in substantial ways. At the most advanced dimensions, the difference between a high-performance, energy-efficient device and one that falls short can come down to just a few atoms.
This matters because every major AI system ultimately depends on advanced logic chips that deliver more performance without a proportional increase in power consumption. As AI systems demand more capability within tighter power envelopes, controlling atomic-scale variation becomes critical.
Materials innovation dominates geometric scaling
Future logic devices require entirely new combinations of materials across the transistor stack, from channel and source/drain epitaxy to high-k dielectrics, dipoles, contacts, gap-fill structures and area-selective films. Success no longer depends solely on shrinking and thinning incumbent films, but on discovering and engineering new materials to deliver the required physical and electrical properties. Devices must switch faster, consume less energy, achieve the right capacitance and resistance characteristics, and display remarkable consistency across hundreds of billions of transistors on every advanced chip, as dimensions continue to shrink.
Perhaps the greatest challenge is control. As GAA technology evolves toward future CFET architectures, every layer must be deposited with extraordinary precision and repeatability. Conformality, selective deposition, gap-fill performance and atomistic control of channel thickness across all devices become critical requirements for continued scaling.
This is where our expertise comes together
Advanced logic is no longer only about architecture and shrinking existing films. It is about engineering the right materials, delivering the right physical, chemical and electrical behavior, and controlling every layer with atomic precision. These are the challenges ASM has been solving for decades. Through technologies such as ALD and epitaxy, we enable the material combinations, device performance and atomistic control required for next-generation logic devices. Looking ahead, future GAA nodes and the eventual transition to CFET are expected to further increase this intense focus on materials, as scaling becomes increasingly dependent on precision materials engineering.
The AI hypercycle may be measured in trillions of operations per second, but its future will increasingly be determined on a far smaller scale. Atoms. Angstroms. And the ability to control them with extraordinary precision.
