Silicon chip giants Intel Corp. and Micron Technology Inc. have unveiled a totally new kind of memory technology called 3D Xpoint that they say offers up to 1,000 times the performance and endurance of regular NAND flash.
The two partners are so excited that Rob Cooke, a Senior VP at Intel charge of non-volatile memory, called it a “breakthrough in technology”, adding that he views it as a “new class of memory because it’s not a little faster, it’s a lot faster, it has a lot higher endurance.”
Announced on Tuesday, Intel and Micron said the 3D XPoint non-volatile memory tech is ten times denser than regular memory chips, adding that the design represents a brand new class of memory chips for the first time since NAND flash was created almost 25 years ago.
The companies said that 3D Xpoint is targeted at data centers, and described how applications and services would benefit immensely from “fast access to large sets of data.”
According to the partners, the non-volatile memory tech is able to serve as a kind of memory pool for system memory and storage. They’re positioning 3D Xpoint as a combination of the performance, density, power and cost advantages that come with current memory technologies, touting it as being 1,000 times faster than conventional Flash, with 1,000 times the write endurance.
Intel and Micron built 3D XPoint using a process called 3D stacking, which allows thin layers of memory to be stacked on cells to boost their density. The technology has been around for over a decade, but Intel and Micron’s breakthrough came with the development of what they call a “cross point structure” where submicron columns are connected by perpendicular wires, an architecture that allows individual memory cells to be addressed by selecting the top or bottom wire.
The companies said the architecture is like a “three-dimensional checkerboard” wherein they position memory cells at the intersection of word and bit lines. Using this high-density configuration, memory cells can be addressed individually, allowing data to be read and written in small chunks while simultaneously speeding up the read/write process.
That’s not all they’ve done though. In addition, Intel and Micron have created what’s called a “selector” that allows read/write processes to be made without the need for a transistor, further reducing the size and cost of its 3D XPoint chips. Conventional DRAM tech needs a transistor in each and every cell, which adds considerably to the size and cost of each chip. The partners also noted that DRAM tech seems to have hit the proverbial wall as well, because its density is no longer doubling every two years as before.
By combining their new stackable architecture and transistor-less design, Intel and Micon claim they can build chips with up to 10 times the density of conventional volatile memory chips of the same size.
One problem that remains to be overcome is latency, as most data centers remain packed to the rafters. However, the partners claim that 3D XPoint significantly reduces lag time between memory, processing and stored data. This allows more data to be stored closer to the processors, the companies said, boosting the speed it takes to access “enormous data sets” far beyond that of current memory technologies.
Latency with 3D XPoint can be measured in the “10s of nanoseconds”, according to the companies. That compared to the “10s of microseconds” with conventional NAND. It might not sound like a lot, but that kind of performance boost is more than enough to eliminate the bottleneck between stored data sets and processors.
To begin with, Intel and Micron said they would initially target 3D XPoint at high-performance SSD storage, and they’ve promised to begin sampling the technology later this year. Each company is currently developing unspecified products based on 3D XPoint, but there’s no indication yet on when these might be available.