For more than a decade, the NAND industry has followed a simple roadmap:
“Build higher.”
After planar NAND scaling reached its limits around the 16nm to 12nm generation, the industry shifted toward 3D NAND. What followed was an intense competition in layer count.
64 layers became 96.
96 became 176.
176 became 232.
Today, the industry is moving toward 400-layer NAND and beyond.
But a new battle is quietly emerging.
The future of NAND may no longer be determined by how many layers can be stacked.
Instead, it may be defined by:
How efficiently different pieces can be bonded together.
YMTC Saw It Early
Back in 2019, YMTC introduced a different concept.
Instead of building the memory cell array and peripheral circuitry together on the same wafer, YMTC separated them and bonded them together later.
At the time, many industry experts questioned the approach.
The criticism was simple:
- Additional process complexity
- Higher manufacturing cost
- Uncertain economic benefit
As a result, major memory manufacturers such as Samsung and SK hynix did not aggressively pursue this approach.
The prevailing view was:
“Why add cost if the current approach already works?”
At the time, that was a reasonable question.
AI Is Changing the Equation
The rise of AI is creating a very different set of requirements.
Historically NAND was optimized primarily for density and cost.
Today, performance matters more than ever.
AI data centers increasingly demand:
- Higher bandwidth
- Lower latency
- Faster interfaces
- Better power efficiency
This creates a growing challenge.
The optimal process technology for a 300-400 layer NAND cell stack is not necessarily the optimal technology for high-speed peripheral logic.
Trying to optimize both simultaneously is becoming increasingly difficult.
The memory array wants density.
The peripheral circuitry wants speed.
The two objectives often compete with each other.
Separating Cell and Logic
As a result, the industry is gradually moving toward a new architecture.
Build the NAND cell array independently.
Build the peripheral logic independently.
Then bond them together.
In other words:
“Use the best process technology for each function and integrate them later.”
What once seemed economically questionable may become technologically necessary.
Interestingly, some reports suggest that certain aspects of YMTC’s architecture have become important enough that competitors are now willing to license related technologies rather than reinvent everything from scratch.
The Lesson from History
This trend reminds me of discussions I experienced more than twenty years ago while working on FinFET and GAA technologies.
In the early 2000s, it was not uncommon to hear comments such as:
“Stop working on these strange ideas and help improve manufacturing yield.”
At the time, such comments were understandable.
Companies always have immediate problems.
Yield.
Cost.
Schedules.
Production.
Future technologies often look expensive, risky, and unnecessary.
Yet many of the FinFET and GAA concepts developed during that period eventually became industry standards.
Patents that once seemed speculative later became foundational intellectual property supporting multi-billion-dollar businesses.
Why Long-Term Thinking Matters
Semiconductor companies operate in one of the most capital-intensive industries in the world.
But capital alone is not enough.
The real competitive advantage comes from having people willing to think five, ten, or even twenty years ahead.
Every company faces short-term pressures:
- Quarterly earnings
- Yield issues
- Market share battles
- Cost reduction targets
Yet if an organization only focuses on immediate returns, it may miss the next industry transition.
HBM is a perfect example.
Not long ago, many people viewed HBM as a niche product with limited commercial value.
Today, it has become one of the most important technologies in the AI era.
The Role of Leadership
Maintaining long-term research programs is rarely easy.
Many future technologies appear unattractive in their early years because they generate little immediate profit.
Ultimately, sustaining those efforts depends heavily on leadership.
Not every CEO needs to be Elon Musk.
But every technology company needs leaders who are willing to support people looking beyond the next quarter.
Because the technologies that define the future often begin as ideas that seem uneconomic in the present.
Conclusion
The NAND industry has spent the last decade fighting a stacking war.
The next decade may be about bonding.
The industry’s future may depend not only on how many layers can be built, but on how effectively memory cells and peripheral logic can be optimized independently and integrated together.
History repeatedly teaches the same lesson.
The technologies that appear expensive, impractical, or unnecessary today often become the foundations of tomorrow’s industry.
Perhaps NAND bonding technologies are standing at that exact crossroads right now.