Trends in memory chips such as NAND flash and non-volatile memory show the parts are gaining importance in the data center.
Data centers are beginning to transition from CPU-centric systems to data-centric systems.
- a positive for memory makers Micron and Everspin.
- a negative for processor makers Intel and AMD.
The primary message from NAND Flash producers was each company’s efforts toward increasing Gb/area of silicon. Samsung mentioned its plans for a 1Tb device in 2018 and the ability to stack up to 32 in a package. These packages can be used to build 128TB 2.5 inch SSDs or M.2 cards with 16TB. Despite these technology advances, the conversations surrounded the continued shortage of NAND Flash despite the increased CapEx spending. With the spending focused on increasing layers per wafer and not on wafer capacity, we expect that the NAND Flash shortage can continue well into 2018.
Micron has greatly improved its Flash cost structure, and expect the company’s costs will continue to decrease moving forward. The company reiterated that it has the smallest 3D NAND die size in the industry, which allows its current products to be more competitive in terms of cost, as opposed to its competitive positioning in planar NAND where the company had higher costs than the industry average, and therefore lower gross margins. Micron’s smaller 3D NAND die size is the result of its Floating Gate architecture, which was also the standard for planar NAND, as opposed to Charge Trap Layer architecture used by Samsung, Toshiba, and SK Hynix.
Persistent memory solutions include memory bus nvDIMMs based on either battery/capacitor backed DRAM with NAND Flash, MRAM, 3D XPoint and a few other ideas in development including RRAM, PCM and Memristors. Investors should watch these developments. Persistent memory could be as revolutionary to data center performance as SSDs.