At the 2018 SNIA Storage Developers Conference Jim Handy and I gave a talk about the prospects for emerging memory technologies that could eventually replace SRAM, NOR, DRAM and possibly NAND flash. This projection of the future of non-volatile emerging memories drew on material from our Coughlin Associates and Objective Analysis publication: Emerging Memories Poised to Explode: An Emerging Memory Report (https://tomcoughlin.com/tech-papers/).
First, let’s look at what is driving the NAND flash and DRAM markets, since these are the dominant memory technologies currently in use. NAND flash has achieved high volume manufacturing that has driven the price in $/GB of this technology to the point where useful quantities of flash memory are available for many applications, thus displacing an increasing number of HDDs, particularly for client applications. DRAM, unlike flash memory, is a volatile memory and needs frequent refreshing to maintain the information stored on DRAM memory Cells.
Before 2004 DRAM was less expensive in $/GB than flash memory. But because of the expanding consumer applications for flash memory, particularly in mobile phones, in 2004 the price of flash went below the price of DRAM and has continued to fall ever since. By 2016 the shipping volume of NAND flash exceeded that of DRAM.
DRAM and NAND $/GB Price Trends Image from our SDC Presentation
Flash memory, like DRAM, experiences boom and bust years. In boom years there are shortages of chips to support demand, and as a consequence the price of NAND flash chips increases. This leads flash memory companies to invest in new NAND flash capacity. Generally all the flash memory companies are experiencing these shortages at the same time and they invest in new fabs about the same time. When this capacity comes on line the shortages disappear and there is a surplus of NAND flash versus demand, and the price of NAND flash drops. There are no new investments in NAND flash fabs at that point and this state continues until demand grows to outstrip the supply of NAND flash and the cycle starts over again. This has happened several times over the history of the semiconductor industry.
With the migration to 3D from planar NAND nearing completion with new NAND fabs we are entering what appears to be a significant oversupply of NAND flash that will probably last for the next couple of years. At the same time, Chinese NAND flash manufacturers will start contributing to this oversupply by 2020, perhaps extending the oversupply situation for longer than is normal.
The planar NAND plants will likely be converted to DRAM manufacturing, leading to an oversupply in DRAM in the next few years and significant price and profit declines in DRAM as well. Decreasing prices of conventional memory, make it hard for emerging memories to replace them. Furthermore, NAND flash is projected to go from today’s announced 96 layers to over 500 layers over the next few years. This will continue the $/GB price declines for NAND flash into the future.
So where will emerging memories gain a foothold? Well, Everspin is making magnetic random access memory (MRAM) standalone memory chips that are used for caching and buffering in many applications, including SSDs and HDDs. IBM announced an SSD that had 160 MB of MRAM capacity using 256 bit Everspin MRAM chips in August and there are many other storage and other systems that are using a bit of MRAM to improve overall system performance. MRAM is very fast, approaching DRAM speeds, has write and read speeds that are very close to each other and has very high endurance.
All the major foundries have teamed up with MRAM companies to make embedded products using MRAM primarily to replace NOR or SRAM memory (including TSMC, UMC, Samsung and Global Foundries). By next year it will be clear if the market for embedded MRAM is taking off, the way these companies project that it will. MRAM will be particularly useful where power consumption is an issue in IoT devices.
Intel’s Optane memory, to be available in DIMM form for populating computer and server memory buses, will be available in 2019. This is supposed to provide improved performance for Intel servers, giving the company an advantage over AMD and other competing server processor manufacturers. Intel is believed to be selling its Optane products at a discount to the actual production costs in order to keep the prices below DRAM and thus drive adoption of the technology. With possible declines in DRAM prices in the next few years it will be interesting to see if Intel will continue this approach.
Besides MRAM and Intel’s Optane there are other emerging memories such as resistive RAM, phase change memory, ferroelectric RAM and carbon nanotubes that could also compete against current memories used in embedded applications. The standalone market for memory chips will provide some of the projected volume of emerging memories, but embedded applications, particularly for mobile and AI application, will likely be the big drivers for embedded emerging non-volatile memory. Our report projects that overall emerging NVM market could exceed $6B by 2023 and generate many exabytes of memory production by 2028 as shown below.
Emerging Memory Low, Baseline and High Capacity Projections Image from our SDC Presentation
Emerging memories will join conventional memories such as NAND flash and DRAM in new applications and electronic systems. These products are likely to replace NOR and SRAM initially and only impact some DRAM closer to the end of the next decade.