Memory Chips in Data Centers: Something's Gotta Give
Some of the biggest semiconductor companies in the world are in a race to build a new generation of memory chips, and the stakes couldn't be higher -- not for the chip makers themselves, nor for their data center customers.
Progress in network virtualization depends on data centers that push every performance metric as hard as they can, and to that end data center operators want storage ICs that are faster than the flash memory (a.k.a. NAND).
To address that need, Intel Corp. (Nasdaq: INTC) and partner Micron Technology Inc. (Nasdaq: MU) are developing what they call 3D XPoint memory. There's a commonly held assumption they have a market lead.
Samsung Corp. subsequently announced it has a variant of its 3D NAND in the works that it is calling Z-NAND.
The Intel/Micron team and Samsung are racing Crossbar , SK Hynix Semiconductor Inc. , Panasonic Corp. (NYSE: PC) and the team of Hewlett Packard Enterprise annd SanDisk Corp. (Nasdaq: SNDK), which are pursuing yet another type of memory device known as resistive random access memory (a.k.a. ReRAM or RRAM). Some of them have been working on ReRAM for six years.
Meanwhile, most of the companies that build flash memory -- including some of those pursuing alternatives to straight-up flash -- are looking to improve their flash products.
Let's back up a bit: Inherent in the notion of virtualization is moving processing into the cloud -- into data centers. A natural consequence of that shift is that internal traffic within data centers (so-called east-west traffic) has increased at a rate far, far higher than in the Internet in general.
That's why it is data centers that are driving the evolution of Ethernet networking from 10G to 25G, 50G and 100G, and that's why it's data centers that are among the key drivers of innovation in higher-end CPUs.
From the data center standpoint, faster and smarter interconnect and processors are all well and good. Improved interconnect gets data from processor to processor faster, and advanced processors crunch data that much faster. It will never be enough, however, not while data center traffic keeps growing so rapidly.
So where to eke out more improvement? One of the last processes yet to be accelerated is fetching data from memory.
Flash was a compromise from the get-go. Everyone wants to replace disks, which are highly prone to failure and take a lot of energy to run, significantly more than solid state memory.
DRAMs are very fast, but they are also volatile, meaning they constantly need to be refreshed else the data they contain is lost; that also takes a lot of energy. Flash is significantly slower than DRAM, but on the other hand it is nonvolatile; no refresh is required. Those properties make flash a good, though not ideal, replacement for disk drives almost everywhere, including in data centers.
Hence all the work to devise faster nonvolatile memory chips, such as 3D XPoint, Z-NAND and ReRAM.
here on Light Reading.
Mellanox Technologies Ltd. (Nasdaq: MLNX) is among the leading suppliers of data center interconnect, including a new class of intelligent interconnect designed to ease the flow of east-west traffic, which is a long way of saying that Mellanox has a ringside seat for the competition among memory vendors as they try to devise faster alternatives to flash.
But there's that third option most people keep overlooking. Mellanox thinks flash is going to last longer than most people think, according to Vice President of Marketing Kevin Deierling.
"Right now, flash looks slow," Deierling said, "but the flash guys said, 'Oh, you need faster flash? We can do that.' " The key parameter is read/write latency, and Deierling said there is no reason that flash manufacturers can't improve that one way or another.
Meanwhile, Intel and Micron seem to be taking a bit longer than anticipated with 3D XPoint, he noted.
In terms of memory speed, you've got DRAM at one end of the spectrum and flash on the other end, and the industry is trying to devise something that falls in between them, Deierling observed. But if the companies working with standard flash can improve them enough, the gap narrows -- possibly enough to eliminate the opportunity for the newer, more exotic variants.
Would that kill 3D XPoint or Z-NAND or ReRAM? Almost certainly not. There are other verticals that will welcome any improvement; gaming is one. But there are few verticals that buy leading-edge memory in volumes that get close to matching the data center market.
Violin, Mellanox believes, will be only the first of the all-flash array vendors "that will finally give up the ghost. The competition will get fierce as the big boys and especially the now colossal Dell-EMC hit their stride. It will become increasingly difficult for the smaller guys -- and maybe even some of the bigger guys -- to compete. Consolidations and pink slips will be the order of the day in 2017."
— Brian Santo, Senior Editor, Components, T&M, Light Reading