In this article I'm going to share some approximate numbers on certain semiconductor demand and available supply. Since I will be primarily concerned with Intel (INTC) and Micron (MU) the numbers will be for Application Processors (AP) and DRAM and NAND memory.
We know that the A7 AP made by Samsung (OTC:SSNLF) for Apple (AAPL) is a billion transistor device with a size of 100 Sq. mm. We know that the Intel 22 and 14nm devices for smartphones and tablets will be smaller than that, but for the purposes of this piece we will be satisfied with 100 Sq. mm for AP.
We know that a 4Gb DRAM at current technology is between 60 and 80 Sq. mm depending on whether the application is PC or mobile, so I'm going to use an average of 70 Sq. mm for DRAM. The smallest cell and largest capacity NAND memory is the Micron 16 nm 128Gb 2D NAND chip and it has an approximate size of 150Sq. mm. We think there will be 1.2 billion smartphones shipped in 2014, some smarter than others. They will have an AP and probably average 1GB of DRAM and 16GB of NAND. We think there will be 300 million tablet computers shipped in 2014. They will have an AP and probably average 2GB of DRAM and 64GB of NAND.
We are quite certain that 300 million PCs of one configuration or another will ship in 2014. They will have an Intel CPU of about $110 average price and about 160Sq. mm in size. The memory will average about 5GB of DRAM (10 4Gb chips) and (here's the point of the article) either a hard disk drive (HDD) or a solid state drive (SSD). For the purposes of this piece, I'm going to assume 100 million will be SSD at an average capacity of 192GB (12 128Gb chips)
|Device||Processor size Sq. mm||DRAM size Sq. mm||NAND size Sq. mm||Total silicon per device Sq. mm|
|PC||160||700||1800 X .33 =600||1450 - 2660 with SSD|
Next we will figure the number of 300mm wafers necessary for each market segment, the assumption is 58,000 net usable Sq.mm per 300mm wafer:
- Smartphones - 58,000/390 = 149 per wafer. 1.2 billion/ 149 = ~8 million wafers (2.4 million for AP alone)
- Tablets - 58,000/980 = 59 per wafer. 300 million/59 = ~5 million wafers (600,000 for AP alone)
- PCs - 58,000/1450 =40 per wafer. 300 million/40 = ~7.5million wafers (1 million wafers for CPU alone)
- NAND for 100 million 192GB SSD = 3.12 million wafers (9.36 million for 300 million SSD)
The biggest single user of silicon is not processors, it is DRAM, with NAND memory poised to pass DRAM like it was stopped (which it is). Even if SSD usage in PCs did not explode, NAND for all applications is the biggest consumer of silicon wafers.
SSD are so superior to HDD that with a slight marketing nudge the demand for SSD could absolutely swamp the ability to make enough NAND memory to serve the market. The NAND memory market is in approximate balance today. To move one third of the PC to SSD would require about 3.12 million new wafers. The biggest fabs in the world are capable of about 1.2 million wafers per year. That means the inevitable move to SSD will require about three new giant fabs per year to get there or about nine new fabs to fully convert PC HDD to SSD. Each of those fabs will cost $5 billion plus.
Who will start it?
Gartner apparently doesn't believe there is enough NAND capacity to get SSD moving; I guess they haven't been following Intel. Intel has two of these giant fabs in Chandler, AZ and Hillsboro, OR, respectively referred to as "FAB 42" and "FAB D1X". Combined, these fabs would probably be capable of 3.4 million 300mm equivalent wafers per year. Combined, these two new Intel fabs could take the first step in converting the PC business from HDD to SSD.
For Intel to re-enter the memory business with the right gross margins they would likely have a much lower cost-by-design NVM technology in their toolbox and would spring it on the world much as they did with 22nm TriGate transistors. Intel has a remarkable ability to keep a secret.
During the recent Intel investor meeting, CFO Stacy Smith said that Intel was at 80% capacity. Since neither of these fabs are operating yet, Stacy could be telling the truth. As soon as those giants process a single wafer they will have to be included in the total capacity number to calculate an honest utilization rate. That would bring the utilization rate down to 40-50% and be very hard for Stacy to explain in the next earnings conference call.
Every bit of the smartphone and tablet business would only require 3 million wafers per year and foundry work is a tiny, tiny fraction of that. There is no target market existing today that could absorb the capacity represented by those two huge new and unused Intel fabs EXCEPT memory and specifically NAND memory for SSD.
To properly serve the mobile business, Intel will need to acquire DRAM capacity since the mobile AP and DRAM are packaged together. Without reliable internal access to DRAM, Intel will be at a perpetual disadvantage to Samsung as a supplier of mobile solutions. They could build or buy the DRAM part. I have been convinced that Intel will buy Micron to solve that problem.
The problem with that is that if Intel announced a NAND initiative BEFORE they bought Micron, ALL memory companies would experience a severe decline in stock prices. Intel could still buy Micron alright, but at a much lower price than they appear to be worth at the moment. With the Intel "Black Swan" circling above Micron (and all memory companies), I have decided the prudent thing to do is to move my options play out of Micron and over to Intel.
I believe Intel will move into the future $50-70 billion SSD market with cannons blazing…when the time is right. If the SSD initiative is not revealed in the next 3-6 months it is not likely to happen at all and my speculation is wrong.
Intel has the capacity. Others would take 2-3 years to build capacity. Intel needs to occupy those new fabs soon. During the recent Intel Investor meeting a SVP and an EVP both made strange "loving" comments about non-volatile memory. Connecting the dots, I think Intel will target the SSD business with a technical and cost advantage and the other NAND players will suffer collateral damage in the process. Therefore, I feel Intel is the better investment vehicle at the moment.