Memory and storage technologies such as DRAM and NAND have been around for decades, with their original implementations able to perform only at a fraction of the level achieved by today’s latest products. But those performance gains, like most in computing, are typically evolutionary, with each generation incrementally faster and more cost effective than the one preceding it. Quantum leaps in performance often come from completely new or radically different ways of solving a particular problem. The 3D XPoint technology announced by Intel in partnership with Micron comes from the latter approach.
“This has no predecessor and there was nothing to base it on,” said Al Fazio, Intel senior fellow and director of Memory Technology Development. “It’s new materials, new process architecture, new design, new testing. We’re going into some existing applications, but it’s really intended to completely evolve how it’s used in computing.”
Touted as the biggest memory breakthrough since the introduction of NAND in 1989, 3D XPoint is a new memory technology that is non-volatile like NAND memory, but is up to 1,000 times faster, with a faster speed only attainable by DRAM, and with endurance up to 1,000 times better than NAND.
3D XPoint owes its performance attributes and namesake to a transistor-less three-dimensional circuit of columnar memory cells and selectors connected by horizontal wires. This “cross point” checkerboard structure allows memory cells to be addressed individually. This structure enables data to be written and read in smaller blocks than NAND, resulting in faster and more efficient read/write processes.
Removing bottlenecks in a system is a key method to increase overall performance. Memory in particular has been a growing barrier, primarily because consistent performance gains in processors in recent years have dramatically outpaced both the speed of hard disks and the cost and density of DRAM.
“What’s exciting about the technology is that it unleashes the microprocessor. It gets more data closer to the CPU. It has 10 times the density of DRAM at near levels of performance, and it allows people running applications to have much more data available to them,” said Rob Crooke, vice president and general manager of Intel’s Non-Volatile Memory Solutions group. “Conversely in the storage, it’s up to 1000 times faster than NAND. To put that in perspective, most people have experienced an SSD versus a hard disk, where the SSD is about 1,000 times faster than a hard disk. This new technology is going to be that same level of pop, like everything’s in memory.”
“We’ve looked at gaming performance, and it has a phenomenal impact and gives the game creators much more freedom,” explained Crooke. “As opposed to constraining their game levels to how much they can fit in memory and then loading a new level, they now have total freedom to create a much richer game experience and one that’s seamless and continuous, and they can decide if they want to break it up. It’s at their artistic and creative discretion to do that, as opposed to some physical limit like memory size.”
“It’ll be a game-changing experience not only in the client platforms, but also in the data center where they’re trying to analyze remarkable amounts of big data,” Crooke continued. “More and more data needs to be driven to the CPU to analyze faster. Having much more data available to the CPU at a very short latency is pretty exciting.”
Micron and Intel have been jointly developing this technology since 2012. There was, however, basic research on various technologies at both companies for years prior to this partnership. The research team could have tried an easier route—committing to performance and density, or performance and cost, for example—but “if you want to change something, you’ve really got to go for that tougher problem and tie them all together,” Fazio explained.
In 2012, Micron and Intel agreed to jointly pursue the most promising technologies from the research findings. Hundreds of Intel and Micron engineers have been involved in developing the technology to its current state, spanning facilities in California, Idaho and around the world. Over the last three years, the process development for this technology occurred in Micron’s state of the art 300mm R&D facility in Boise, Idaho.
“Nobody has ever attempted productizing a stackable cross point architecture at these densities. Learning the characteristics and developing the integration methods for this novel architecture was full of engineering challenges,” said Scott DeBoer, vice president of R&D at Micron. “3D XPoint technology required the development of a number of innovative films and materials, some of which have never before been used in semiconductor manufacturing. Understanding the characteristics and sensitivities of these new materials and how to enable them was daunting.”
3D XPoint, NAND and DRAM
While 3D XPoint may have capabilities that can displace DRAM and NAND, DeBoer noted that it’s an additive technology that will co-exist with current solutions while also enabling new innovations. “DRAM will still be the best technology for most demanding highest performance applications, where non-volatility, cost and capacity are less critical. 3D NAND will still be the best technology for absolute lowest cost, where performance metrics are less critical.”
What could be a significant factor in these different memory solutions co-existing is that they can all share manufacturing facilities. “This technology is fabricated using the same manufacturing lines and methods as conventional memory technologies,” said DeBoer. “With the cross point architecture and the materials systems required for the new cell technology, some unique tooling was developed, but these requirements are on par with standard technology node introductions for NAND or DRAM. This technology is fully compatible and not disruptive to current manufacturing lines.”
Scalable Into the Future
The future of this technology looks wide-open too. “The cross point memory cell should be the most scalable architecture,” said Crooke. “It should allow us to scale the memory technology to pretty good densities yet allow it to be byte-addressable or word-addressable like memory is, as opposed to NAND, which is accessed in blocks of data.”
“Because it does not require the overhead of additional access or select transistors, the stackable cross point architecture enables the most aggressive physical scaling of array densities available,” DeBoer added.
A technological solution that paves the way for new models of computing doesn’t come by very often. It took teams of hundreds of experts, countless flights, and constant open lines of communication and cooperation to bring make 3D XPoint technology possible.
“Micron and Intel have a long working history inside our NAND JDP and our IMFT joint venture. This made enabling the team cooperation and performance that much easier as we have already strengthened and grown the partnership in that program,” said DeBoer. “Entirely new technologies don’t come around very often, and to be part of this team was truly a once-in-a-career opportunity.”
“One of the things that we should be proud of is the persistence we’ve had over a long period of time,” Crooke added. “Working on a technology problem that you don’t know is solvable, for a sustained period of time, requires a level of confidence and stick-to-it-iveness.”
This content was originally published on the Intel Free Press website.