The atomic path to artificial general intelligence
How atomic-scale precision is advancing logic and memory for AI
At SEMICON Taiwan, Abhudaya Mishra and Angada Sachid shared complementary perspectives on materials innovation, highlighting why breakthroughs in materials, not just device shrinkage, are now driving the next wave of AI progress.
AI is advancing toward increasingly ambitious milestones, from today's inference and agentic systems toward the long-term goal of artificial general intelligence (AGI), AI capable of performing at or beyond human levels across virtually any task. Reaching that goal requires scale: more advanced logic and memory, more advanced packaging, and more capable fabs, stacked into ever-larger AI systems.
However, scale comes at a cost. The energy demands of next-generation AI systems are rising sharply, which is pushing the industry to solve for efficiency as much as it solves for performance. Adding transistors is no longer enough. One of the easiest ways to improve efficiency is bringing memory and compute closer together, which is why the industry is moving toward 3D packaging, heterogeneous integration, and hybrid bonding.
Abhudaya Mishra, Corporate Director, Technology, Innovation, and Market Research Organization
"AI may be capturing the headlines, but what's fueling its advancement is far less visible. At the foundation of the world's most advanced AI are breakthroughs in materials science, engineered atom by atom."
As shrinking slows, materials take on a greater role
For six decades, making transistors smaller, including the shift to gate-all-around (GAA) and complementary FET (CFET) architectures, drove most of the industry's performance gains. AI workloads now require performance gains that geometry alone can no longer provide. As a result, the industry is turning to system-level co-optimization and new engineered materials to keep performance, power, area, and cost (PPAC) on track.
In turn, atomic-scale precision has become a key part of the equation. A difference of one or two atoms can affect whether a chip is power-efficient or power-hungry. Two deposition technologies are central to controlling this: Atomic Layer Deposition (ALD) and epitaxy.
Angada Sachid, Senior Director, Technology, Innovation, and Market Research Organization
"We're stacking hundreds of these atomically precise layers on top of each other. That's the level of monolayer control the next generation of memory will demand."
Atomic-level control in practice
ALD has earned its place at the center of this transition through three successive waves of innovation, each one adding new capability.
- Thermal ALD (tALD): introduced high-k metal gate transistors, a foundational shift in how transistors were built.
- Plasma-enhanced ALD (PEALD): enabled low-temperature SiO deposition for double patterning.
- Area-Selective Deposition (ASD): brought fully self-aligned vias and area-selective processes into high-volume manufacturing.
The range of materials ALD can deposit has expanded just as steadily, from a handful of elements decades ago to >80% of the periodic table today.
ALD is uniquely different from other deposition technologies in terms of directionality and film-growth mechanism, making it ideal for angstrom era nodes. ALD deposition builds films through self-limiting surface reactions, one atomic layer at a time, allowing it to conformally and atomistically coat even the most complex 3D structures.
This kind of atomic-level control isn't limited to deposition. Speaking at SEMICON Taiwan, Angada Sachid pointed to epitaxy's growing role in future memory, as 3D DRAM moves toward stacking hundreds of layers to increase density.
