WFE Intensity Step-up: 3D NAND Case Study

After declining significantly from 2000 to 2015, we estimate that the NAND wafer fab equipment (WFE) intensity trendline took a 10-point step higher-from around 15% to around 25%-driven by the 2D to 3D NAND inflection. NAND WFE intensity has remained sustainably higher ever since. Let's examine the factors that led to this increase.

3D NAND Case Study

Our analysis suggests that from 2000 to 2015, NAND WFE intensity decreased by nearly 30 points, from around 50% to around 20% (represented by the dotted line in Chart 1). While NAND revenues increased, annualized NAND bit growth gradually declined over this period owing to the deceleration of classic 2D scaling, which is consistent with our Battle of Exponentials framework. Toward the end of this period, NAND makers began a transition from 2D to 3D architectures. We estimate that the industry's transition to 3D (represented by the shaded area in Chart 1) progressed from less than 10% of NAND capacity in 2015 to greater than 80% in 2021 as the new architecture permeated most NAND applications.

2D to 3D Transition = Higher WFE Intensity

As I wrote in my previous blog post, WFE intensity is driven by capacity and complexity. The latter often triggers the former. 3D NAND introduces an increasing number of new steps (complexity) compared to 2D NAND. Accordingly, NAND makers built new (a.k.a. greenfield) fabs suited to the new technology to demonstrate viability prior to converting 2D fabs to 3D, which they did later.

From Exponential Moore's Law Scaling to Decelerating 3D Scaling

3D NAND scaling is different in another important way. While successive generations of 2D NAND scaling generated exponential bit growth, each increase in a defined number of 3D NAND layers results in progressively slower bit growth. Simply put, while the transition from 32L to 64L TLC devices produced 100% bit growth, the 64L to 96L transition produced only 50% bit growth, and the 96L to 128L transition produced only 33% bit growth. As percent bit growth moderates, manufacturers either need to accelerate the number of layers and steps, or they need to add wafer capacity. Either way, NAND WFE is sustainably higher.

Chart 2 illustrates that in the 2015 timeframe, the greenfield capacity cost for a given number of 3D NAND wafer starts was about 1.5X the cost of 2D NAND, and the delta widened to 2.2X by 2021.

In addition to helping us think about NAND WFE intensity, we believe our 3D NAND case study may be instructive as we anticipate the potential impact of future 3D architecture transitions expected in DRAM and Foundry/Logic.

^*Average WFE $/wafer across comparable technologies from multiple NAND makers.