• Positioned as secondary storage, 3D NAND is the unsung hero of AI, and the disruptions it brings provides new entry points for 3D NAND.
• It is critical in supporting edge AI applications as models need to be stored close to processing.
• 3D NAND requirements across a range of metrics are increasing with the rise of AI applications at the edge: higher capacities, performance levels, scale, and the economics for device makers.
• With 1,000-layer NAND the long-term target, cryogenic etching is emerging as a key technology.

Memory has gained further importance over the past few quarters as the shift to GenAI across all sectors accelerates. Embedded and external storage solutions like SRAM, DRAM, and HBM, have emerged as key factors that enable GenAI applications and advanced packaging technologies used to manufacture them. The growing role of memory and storage in enabling inferencing at the edge has been a recurring theme in our research. For insights into why inference at the edge could outpace training in importance, see the blog: ‘Artificial Intelligence and the Edge Debate.’

3D NAND is currently being employed as secondary, high-capacity storage in the GenAI context. It is designed to keep pre-trained models and large datasets close to primary memory and processing units. This proximity supports faster data access and enhances performance. Counterpoint had recently held an interview with Lam Research Group Vice President and General Manager, Harmeet Singh, discussing the significance of 3D NAND.

• Our Whitepaper: ‘Scaling to 1,000-layer 3D NAND in the AI Era’

• Deep dive into the topic with our Lam Research interview

As AI models grow, 3D NAND's importance in supporting complex applications will become even more critical. Counterpoint expects the overall NAND FLASH memory market to more than double to reach $93 billion by 2030, up from $40 billion in 2023.

This massive growth comes with a strong need for NAND Flash makers to deliver on much higher levels of performance, increased yield, and reduced costs; delivering the required tools and solutions to address manufacturing challenges is now critical as complexity increases with the 3D NAND stack growing taller with each successive generation.

Lam Research is spearheading the development and implementation of 3D NAND technology through its cryogenic etching process that will support long-term on-device AI requirements.

In a recent conversation with Counterpoint, Lam Research Group Vice President and General Manager, Harmeet Singh, explained, “We are in the early stages of the AI boom. The technology is expected to transform the industry, and we are working closely with our customers to help enable their next-generation chips. Whether it is leading-edge logic, DRAM, or NAND, these components are key drivers for the processing power and storage that the AI revolution needs.”
Below are some of the key takeaways from our conversation. The full interview can be accessed here.

For the deep dive on cryogenic etching, which Lam Research expects will help deliver 1,000-layer 3D NAND chips in the future, click to download the ‘Scaling to 1,000-Layer 3D NAND in the AI Era’ Whitepaper.

The Unsung Hero of AI: NAND

While GPUs and DRAM often receive the spotlight in AI advancements, NAND flash memory plays a vital, yet less recognized role. Modern AI-powered devices – from smartphones with edge AI capabilities to autonomous vehicles – rely on high-capacity, reliable NAND storage for model storage and real-time processing.

Singh sums up, “NAND is solid-state storage with no moving parts, offering higher reliability and lower energy consumption.” He further describes how the industry has transitioned from 2D to 3D NAND, allowing more memory capacity by stacking layers. This evolution is essential as the demand for data storage grows alongside on-device AI. More broadly, the move towards increasing bit storage densities is being fueled by announcements from chipmakers about increasing the number of layers to 1,000 before the end of the decade. In comparison, we expect to see a 300+ layer chip to be shipped by the end of 2024.

Pioneering Challenges in Vertical Scaling

AI smartphones, PCs, and enterprise servers along with autonomous vehicles and robotics will be the key end-market drivers for 3D NAND growth and the need for more layers will only increase, bringing further complexity to the manufacturing process.

Singh elaborates, “As we increase the number of layers, the structure's height grows, requiring us to etch deep, narrow holes – about 10 microns deep and only 100 nanometers in diameter. This precision is akin to drilling a hole a thousandth the width of the human hair.”

Chip makers are adopting different ways, including leveraging innovative tools and hybrid bonding technology, to overcome the capacity and performance challenges to reduce manufacturing costs and yield losses.

Finding the right balance between bit density, read and write speeds, power, reliability and cost is crucial for applications. Enabling technologies to address 1000-layer scaling in NAND will require innovations across all the three vectors of scaling.

To address these complexities, Lam Research has introduced the cryogenic etching technology.

Singh commented, “Starting in 2019, we developed our cryogenic etching capabilities and have advanced to Cryo 3.0, producing nearly perfect etching profiles. We have processed over 5 million wafers with cryogenic etching, establishing our leadership in this space.”

This should help OEMs accelerate the pace of development in 3D NAND with the aim of meeting increased capacity and bit growth requirements in the long term.

Sustainability and the Future

Sustainability has also been a priority for the company. “Our latest Cryo 3.0 chemistry has significantly reduced energy consumption and the global warming potential of our processes,” Singh said. He further highlighted how pulse plasma technology and innovative chemistries contribute to a more sustainable semiconductor industry.

The new process involving low temperatures coupled with lean chemistry offers an additional window for improvement for NAND manufacturers. Innovations with such precision and uniformity along with delivering lower environmental impact are needed to overcome the AI era's key NAND manufacturing hurdles especially when employed in high-volume production.

A Self-sustaining Cycle: AI for AI Chipmaking

Touching upon Lam Research’s use of AI for optimizing chip production, Singh said, “We are leveraging AI to accelerate the development of our processes by integrating big data, machine learning, and first-principal models. This helps us overcome the increasing challenges in semiconductor manufacturing.”