Disruptive 3D XPoint Technology comes to the Rescue of the Information Age
by Tim Reyes
On July 29th, Intel Vice President of the Non-Volatile Solutions Group, Rob Crooke and Micron CEO Mark Durcan did not mince words. The two described a “revolutionary” and “disruptive” technology and expressed confidence of first delivery in 2016. Durcan called 3D Xpoint (Crosspoint) Technology the first new memory type in the last 25 years (Nand Flash, 1989).
Xpoint is an imminent step function advance for the electronics age that has otherwise followed the more civil Moore’s Law — doubling the number of transistors on an integrated circuit every two years. Crooke stated that Xpoint makes possible today technology not expected for another decade. New applications — yes — but Xpoint is a solution for the demands of the Information Age and the fueling of even faster growth.
How quickly and in which ways will it disrupt the computer industry? Will Xpoint produce a 10 year short hop to future technology? What does it mean for some mainstay technologies of present computer designs?
Consider the performance comparison first. 3D Xpoint is 1000 times faster, 1000 times more durable, and ten times denser than Nand Flash memory. Consider where this is heading. This means the replacement of Nand Flash memory and mechanical hard drives by 3D Xpoint; not immediately, but the writing’s on the wall.
Each memory cell of Nand flash memory that consumers depend upon in, for example, smartphones, USB flash drives or as storage in their laptops, is constructed of floating-gate transistors. Neither Intel or Micron have described the materials involved but Xpoint memory bits do not require transistors. All the material of a Xpoint memory cell is used and a property change occurs when a charge is applied through the Crosspoint lines intersecting above and below each memory cell of the architecture. Xpoint is addressable at the bit level in contrast to Nand Flash which requires block or page read and writes.
As great as this announcement sounds, caution remains among industry experts and engineers. HP and the memristor is the cautionary tale that many recall. In 1971, Leon Chua, defined the term memristor — memory resistor — as a missing link in the set of basic circuit elements. In 2008, Hewlett-Packard announced the development of a memristor that would deliver the same performance increases just announced for Xpoint. In 2011, they projected a production date in 18 months (2013) and now this date stands at 2018 while recent scientific publications are stating that the memristor is an unfeasible technology altogether.
In contrast to HP’s journey, Intel and Micron stated that existing manufacturing processes can be adapted to produce Xpoint, that facilities will not have to be scrapped altogether. The Micron CEO stated that unlike past claims, without mentioning names, their technology was on stage and with manufacturing preparation under way.
What should further quell concerns is Intel’s announcement of Intel Optane Technology at its developer forum in San Francisco, on August 18th that will use Xpoint to deliver “a new line of high-endurance, high-performance Intel SSDs beginning in 2016.”
Intel’s Crooke stated that Xpoint has been under development for 10 years but their collaboration began in 2012. Both executives emphasized they intend to independently build Xpoint products and both will apply their vast array of facilities to produce Xpoint. Licensing the technology is not in their cards for now.
Cross-point Connect is the architecture of Xpoint. The name and the illustrations presented indicate similarities to the Crossbar Latch architecture that was invented by HP in 2001 and was the intended architecture of the memristor. Both architectures are stackable and with near zero dependence on the transistor. But Xpoint is a reality. Considering the development period of Xpoint, both companies built upon the concept of the Crossbar Latch and the discovery of new materials with phase change properties that alters their electrical resistance. One might imagine that without the monkey on their back, i.e. making the memristor commercially feasible, Intel and Micron were able to work more freely and allow the materials to sculpt or set constraints that led to a practical commercially viable design.
Xpoint is disruptive technology at a more fundamental level than, for example, the laser printer or smartphone. It is a temporary solution for a never-ending problem. Data growth drives the need for greater computing and network performance which itself produces a faster more rapid growth of data. Greater data processing demands faster, more efficient processors, data storage mechanisms, and the means to move data from storage to processor and back to storage or to human interfaces.
A quick retirement of long established technology is now likely. Nand Flash solid state drives have been replacing the low-end of storage, primarily for laptops. Intel’s Xpoint Optane solid state drives will be based on a 128 gigabit die from their initial manufacturing process.
Intel’s Crooke during Q&A stated that the initial cost of Xpoint stands between that of DRAM and Nand Flash. Durcan also stated that 3D Nand Flash will continue forward unimpeded by their new technology. They believe that there is no other company with technology that will rival Xpoint any time soon. This seems to indicate that once other developers determine exactly how Xpoint works, they will find a unique process well protected by patents.
So initially Xpoint will not impact Nand Flash production but function as a new memory class as the executives emphasized. However, new architectures such as Optane will begin to replace Nand Flash and conventional hard drives. Once Xpoint is delivered to other manufacturers, new architectures will be developed that improves the performance of Xpoint-based storage and memory devices.
One should also keep in mind that it is the initial cost of Xpoint that stands between DRAM and Nand Flash. With production refinements and increased volume, the price will drop. One only needs to recall the price drops of flash memory cards in this century to imagine the potential for cost reductions. Or recall how the cost of hard drive storage fell in the last 30 years.
A continuing study of hard drive durability by the company Backblaze would indicate that a hard drive replacement rate of about 10% per year is common across all the millions of data servers that Google, Amazon and other corporations are managing. About half of the millions of hard drives in data servers will be replaced in five years. However, within the next five years, Xpoint storage cost is likely to match and exceed hard drives. Already the described performance far exceeds both mechanical and Nand Flash drives. Xpoint durability will significantly reduce the cost of maintenance and replacement of hard drives as applied to vast storage systems. In personal computers, mechanical hard drives will disappear even sooner.
Remember the floppy disk? Soon the 30 year reign of the hard drive will end. The mechanical hard drive and DRAM-based memory are like living fossils still surviving within our computing systems today. DRAM will continue as a staple of computer designs for longer but Intel’s Crooke did refer to consumer products in which Xpoint will completely replace both DRAM and Nand Flash thereby simplifying designs — reducing manufacturing costs and making smaller lighter weight consumer products.
The intense computing demand that is placed on storage servers and supercomputers will still require the use of DRAM for now but it is apparent that the initial performance specs of Xpoint is suitable for all the memory and storage needs of consumer products such as smartphones, or wrist-based devices, touchpads and some classes of laptops. In personal computing devices, DRAM will persist longest embedded as larger memory caches on GPU and CPU chips. Intel anticipates that gaming computers will soon become 8K-capable with Xpoint integrated into GPUs.
Since the announcement, the reaction in Silicon Valley is tempered. There is a cautionary tale of the memristor. There are also other reactions. Many products will not be affected by sudden performance increases. Its not disruptive technology for related developers. Other corporations large and small have silently pulled engineers off their segways and fitness balls to brainstorm and determine how they can create new products or improve existing. There are also many that despite being employed in related engineering fields still have a hard time conceiving what this sudden 10x and 1000x increase in performance means.
Rather than retire complete lines of products, a new line of Xpoint computers — laptop and desktops — are likely to arrive that stand alongside conventional designs. As before, such products lines will be premium priced.
Xpoint will lead to some remarkable personal computers. Motherboards will shrink and altogether, the weight of laptops will take another shaving and see reduced power demands through replacement and simplification; even longer battery life. Xpoint will mark the end of an era of personal computing machines as we’ve known them and have had to manage them — similar to how your father and grandfather had to manage their automobile compared to the vehicles of today.
Standard storage on personal computers is likely to increase quickly to 10 terabytes. Xpoint personal computers will store multiple backup images without hesitation or disturbance to the end-user. All computers heretofore will just shut on and off like a light switch, reboots will be uncommon. Loss of data will become very uncommon due to nonvolatile everything.
While Intel and Micron have described Xpoint as a new class of memory between DRAM and Nand Flash, the adopters will determine for how long Nand Flash co-exists with Xpoint. Also, this announcement represents a first generation of Xpoint technology and its initial performance is already close to DRAM.
The second or third generation could achieve parity or exceed DRAM performance in every way and Xpoint could become the one and only memory type in all computer architectures. This may not even be necessary. New computer architectures could utilize the full potential of Xpoint so markedly that the overall computing performance exceeds existing designs and thereby renders conventional designs obsolete. Game over for DIMM memory, Nand flash, hard drives and architectures of present day motherboards.
Durcan described performance gains for gaming and pattern recognition that will eliminate any delays commonly caused by image data transfers. Today’s top-end performance of AMD and NVIDIA graphics could overnight become the low-end of the performance spectrum. Graphical performance increases should open up new possibilities in the gaming industry and broader adoption of real-time player interaction through virtual reality.
The increased storage capacity of Xpoint will permit larger detailed imagery and mapping information of cityscapes on board self-driving cars. Such cars will use on board data without heavy dependence on the internet to evaluate the near environment. This may permit pattern recognition via stereo imaging to outperform that of laser rangefinding such as is used by Google in their autonomous vehicle.
The announcement emphasized the importance of Xpoint on Enterprise computing. Everything about Xpoint benefits large banks of data servers. Rapid increase in production and lower cost of Xpoint would serve them well. Xpoint will permit the servers of the world’s data to keep pace with demand and also lead to even greater demand.
As CEO Durcan presented their new product he emphasized how by 2020 the world will produce 44 Zettabytes (44 billion terabytes, 1 terabyte = 1000 gigabytes) of new data. 42 Zettabytes is an estimate of the storage needed for every word ever uttered by human beings. While 3D Xpoint is literally coming to the rescue of a world driven by information and connectivity of everything — humans and machines, it will place even greater demands on network performance increases.
With sufficient volume production and cost per terabyte, not just hard drives in networked arrays of data servers will be replaced. A more rapid increase in the number of internet nodes (with arrays of data servers) could permit more mirroring of data around the world and provide a partial solution for increasing the performance of the internet.
The July 29th presentation also pointed to the benefits to genome R&D. Other regimes of bioengineering will benefit as well such as the computer-based search for new drugs. The Ebola outbreak in Africa was halted by medical workers on the ground but also by researchers analyzing potential antidotes with computers. Future outbreaks will be met with even faster response times and delivery of solutions.
Weather modeling could see a sudden improvements that increase the fidelity of predications at the ten kilometer scale or less and for global modeling, reducing the time scale and accuracy for future trends. The next generation of U.S. GOES weather satellites will increase image resolution by 4x, temporal coverage by 5x and add more wavelength bands — all demanding more data storage and performance.The same benefits from Xpoint will be seen in such fields as astrophysics and material science.
What are the broader implications to society and the engineering culture? Will Xpoint spawn a new generation of startups just when it is appearing that investment in startups is stagnating? How will Gen-X and Gen-Y (Millennial) professionals need to adapt their skills due to a sudden increase in performance and capacity and impact on old products or new ones?
With increasing concern focused on the divide and concentration of wealth such as discussed in “Capital in the 21st Century”, will the new wealth that’s created lead to more disproportionate or more even distribution of wealth?
Intel states that Xpoint is making new technology possible now that was not expected for another decade. A precaution for all — buckle your seat belts, acceleration of technology is about to experience a sudden increase.