Intel and Micron Technology have jointly developed a new memory architecture that officials say will enable processors to more quickly access massive stores of data, a needed capability in light of such trends as the proliferation of connected devices, cloud computing and analytics.
In a press conference with analysts and journalists July 28, executives with both companies unveiled the 3D XPoint (pronounced “cross point”) technology, calling it the first new memory architecture since the introduction of NAND flash in 1989. The new memory chips—which are in production now at a joint plant in Utah and will become available in 2016—will be up to 1,000 times faster and have 1,000 times more endurance than NAND, and will be 10 times denser than conventional memory.
Like some other memory technologies, 3D XPoint is nonvolatile, which means data can be retained even with the power turned off, effectively enabling storage of the data. It will enable new designs in devices and new applications that can take advantage of the high performance, speed and density, as well as the lower costs, according to Micron CEO Mark Durcan.
Durcan noted that other vendors have talked for years about creating a new memory technology to address the current challenges, but that Micron and Intel are the first to start production and will be the first to bring it to market.
“This is real technology,” he said during the press conference. “It’s not a PowerPoint presentation.”
Intel and Micron have partnered on memory products over the past 10 years—including introducing 3D NAND products earlier this year—and while engineers from both companies have talked about the need for new storage and memory technologies during that time, they have been working on what would become 3D XPoint since 2012.
There were a number of challenges in developing the new technology, and there was no guarantee it could be done, according to Rob Crooke, senior vice president and general manager of Intel’s Non-Volatile Memory Solutions Group. Engineers from both companies persevered despite understanding the hurdles involved.
“It’s one thing to pursue a problem and solve a problem,” Crooke said. “It’s another thing to pursue a problem over a long period of time that you’re not sure is solvable.”
The key challenge was finding a way to reduce the time it takes for the processor to access the data, an increasingly important factor in the era of big data analytics. The two executives noted that the amount of digital content being created will grow from 4.4 zettabytes in 2013 to 44 zettabytes by 2020, fueled in large part by cloud computing and the rapidly growing number of connected devices that make up the Internet of things (IoT). Cisco Systems officials expect the number of connected devices worldwide will grow from 25 billion last year to more than 50 billion by 2020.
Businesses are increasingly demanding the ability to take these huge amounts of data, analyze them and turn them into useful information in near real time. The 3D XPoint technology—with the greater performance, speeds and densities and the ability to bring the data closer to the processor—will be able to do it in nanoseconds, according to officials with the vendors. It will help in a broad range of areas, according to Durcan, from high-performance computing (HPC) to high-fidelity pattern recognition to genomics research. Gaming will benefit by enabling faster loading of data, the officials said.
Intel and Micron engineers created a three-dimensional architecture that doesn’t use transistors but rather relies on switches. The executives—who showed off a wafer comprising 3D XPoint chips—likened it to a screen door with fast switches where the metal lines intersect. They declined to talk about the materials they’re using for the technology.
Intel, Micron Unveil New Fast Memory Architecture
Intel’s Crooke said the new technology addresses demands that other memory architectures can’t. They usually offer two of the three attributes needed for current and future workloads—speed, density and nonvolatility. The 3D XPoint technology brings together all three.
“You need it to be dense, fast and nonvolatile to fill that computing gap,” he said.
Rob Enderle, principal analyst with The Enderle Group, said the new technology “creates the foundation for a massive increase in overall system performance in line with what we got when we moved from magnetic drives to flash.”
“This is very significant,” Enderle told eWEEK in an email interview. “It promises performance in line with DRAM with all of the advantages we’d typically connect with Flash. It will create the foundation for systems that significantly outperform what we currently have in market. Far faster analytics at a lower cost, far better real time translation, far more intelligent systems. It has applications that start with handheld devices and move to large systems. It may make HP’s Memristor technology obsolete before it is even brought to market.”
Hewlett-Packard engineers have been working on memristor memory technology for years. Memristors are essentially nonvolatile chips that can operate like both storage and memory for the computer, creating a fast memory technology that also can keep data stored when power is turned off. The idea of memristors has been around as theory for decades, but it wasn’t until late in the last decade that researchers showed such chips could be built.
Memristors are expected to be among a range of technologies—including silicon photonics and custom processors—that will form the foundation of a new server architecture HP is working on, dubbed The Machine. The company is five years away from introducing the systems, but officials are preparing to unveil a single-rack prototype of the system in 2016. However, HP engineers are working through issues with the development of memristors, so the prototype will use traditional DRAM chips.