# Why are NAND Flash Fabs so Huge?

Many readers have probably wondered why NAND flash fabs are so enormous.  Although DRAM fabs used to be the largest, running around 60,000 wafers per month, NAND flash fabs now put that number to shame, running anywhere from 100,000-300,000 wafers per month.  Why are they so huge?

The reason is that you need to run that many wafers to reach the optimum equipment balance.  The equipment must be balanced or some of it will be sitting idle, and with some tools costing \$50 million (immersion scanners) you want to minimize their idle time to the smallest possible number.  I am sure that this is a tough problem, although I have never had to solve it myself.

The most important reason that so much attention is focused on this is that the cost of the wafer depends on the efficiency of the fab.  If you built a \$13 billion NAND flash fab that produced 90,000 wafers per month instead of 100,000 wafers per month, then the amount of investment per wafer would be 10% higher.  That can make a significant difference to a wafer’s cost, since about half of the cost of processing a NAND flash wafer is from capital depreciation.

Envision, as an enormously over-simplified example, that your company has decided to paint all of its wafers.  (Nobody really paints semiconductors, but we’ll just pretend that they do.)  This is done using three kinds of tools (I am making these up): A sander, a painter, and a dryer.

The sander can sand three wafers per hour.  The painter can paint two wafers per hour.  The dryer can dry seven wafers per hour.  What is the minimum number of each tool that will assure that all the tools are being used 100% of the time?

You may have noticed that this problem can be solved by finding the lowest common denominator when adding fractions.  You set it up using the number of hours each tool takes (on average) to process a single wafer: 1/3 of an hour for the sanders, 1/2 for the painters, and 1/7 for the dryers.  (The dryer may actually dry all seven at the same time, but you set the equation up as if it’s one at a time.)  You then express these three fractions using the lowest common denominator so that you get 14/42 for sanders, 21/42 painters, and 6/42 dryers.  The optimum combination of tools, then, is 14 sanders plus 21 painters plus 6 dryers to give you a throughput of 42 wafers per hour.  Any number smaller than 42 will result in idle equipment, pushing up your costs.

So, for this setup, 42 wafers per hour is the most economical number.

For a flash fab equipment capacity matching is considerably more complex with process flows that loop back on themselves, equipment that has significant down time and other factors. Currently the average Flash fab is just over 100,000 wafers per month but even bigger fabs with several hundred thousand wafers per month have been coming on-line and the average is growing every year.

To examine the cost impact of fab size, Scotten Jones, President of IC Knowledge LLC has modeled wafer cost versus fab capacity all the way out to 1,000,000 wafer per month. The chart used for this post’s graphic illustrates wafer cost versus fab capacity and includes the percentage increase in cost versus the 1,000,000 wafers per month fab for smaller fabs.  Click on the chart if you would like to see a larger version.

As can be seen from the figure wafer cost for an 8,000 wafer per month fab is a whopping 50% higher than for 1,000,000 wafers per month fab, 70,000 wafers per month is 5% higher and even at 250,000 wafers per month the wafer cost is 1% higher.

Since DRAM is built using many different tools than those used for NAND flash, the optimum number of starts will differ too.  For DRAM 60,000 starts proves to be an efficient volume.

From the perspective of the chart it’s pretty clear why a fab with 100,000 wafer starts might be a good idea – it’s going to be significantly more cost-effective than a 60,000-start fab.

IC Knowledge sells cost and price models for a variety of semiconductor and MEMS applications.  IC Knowledge’s models can be used to investigate cost versus volume as was done here, capital investment costs, wafer costs for new technologies, selling prices for most semiconductor and MEMS products and many other applications.  IC Knowledge’s customers include many of the largest semiconductor and MEMS companies, equipment and materials companies, analysts, and system companies.

Thanks to Scotten Jones for sharing this enlightening chart.