Intel and Micron this week unveiled a new type of memory they plan to mass produce that is purportedly 1,000 times faster than NAND flash and has 1,000 times the endurance.
One thousand times the endurance would be about one million erase-write cycles, meaning the new memory would last pretty much forever.
By comparison, today's NAND flash lasts for between 3,000 and 10,000 erase-write cycles. With wear-leveling and error correction software, those cycles can be improved upon, but still don't get anywhere near 100,000 cycles.
The new product, 3D XPoint, is essentially a mass storage-class memory that, while slower, is still cheaper to produce than DRAM and vastly faster than NAND. Most importantly, it's non-volatile. So when the power goes off, the data remains intact -- just as it does with NAND flash.
"It's more expensive than NAND and cheaper than DRAM, and faster than NAND but slower than DRAM," said Jim Handy, an analyst with Objective Analysis. "This suits it well to create another layer in the memory hierarchy. That's a hard sell, though."
So pumped are the companies about 3D XPoint that they're claiming it's the first new memory category in more than 25 years. According to research analysts, Intel and Micron are not exaggerating. 3D XPoint will reside between DRAM and NAND flash, able to replace both in some instances in enterprise data centers and, eventually, consumer desktops and laptops.
While they announced that they're entering production, Micron and Intel have also left much of their product a mystery.
For example, they're not disclosing what materials are used to create the memory; there are no specifics on performance because that depends on the end product and its use; and while they expect to begin shipping samples soon, Micron and Intel don't expect to roll out products until 2016.
"So that's a puzzle," Handy said.
Russ Meyer, Micron's director of process integration, said 3D XPoint chips are in production but they're awaiting their final form factor according to what equipment manufacturers will need. The new memory is still about five to eight times slower than DRAM.
"It's not as fast as DRAM, so it's not going to replace it in the most latency-valued applications, but it's much higher density and much lower latency than NAND," he said. "If you compare how much faster SSDs are to hard drives and how much faster 3D XPoint is to conventional NAND, it's kind of the same order of improvement."
3D XPoint will have about the same storage capacity as conventional or 2D planar NAND, Meyer said, but it's about 10 times denser than DRAM. That means Intel and Micron's 3D or stacked NAND will still offer greater densities.
The reason 3D XPoint has such a high endurance compared with flash memory is that the NAND uses charge trap flash semiconductor technology, which "traps" or stores electrons in transistors. Flash inherently wears over time as electrical charges flow through the silicon oxide material. Additionally, NAND cells can trap charges where they don't belong, further adding to degrdation of the flash memory, Meyer said.
Solid-state drive (SSD) makers add wear-leveling software, which more evenly distributes writes across the flash memory so as to not wear out one section of the NAND faster than another. By comparison, 3D XPoint (cross point) architecure doesn't store electrons. "The memory element itself is simply moving between two different resistence states," Meyer said.
Not completely unique
While it may be touted as such, 3D XPoint technology is not entirely new. The premise of cross-point architectures is that it removes the need for bit-storing transistors and instead uses a latticework of wires that use electrical resistance to signify a 1 or a 0.
Cross-point architectures that use resistive random-access memory (ReRAM or RRAM) are based on the memory resistor concept, also called Memristor as far back as 1971. Resistive memories have been nearing production for the past couple of years.
Instead of the transistors used by traditional NAND flash to indicate bits of data, XPoint and ReRAM cells change their physical properties to either have a high or low electrical resistance. When in a high resistance state, meaning electricity cannot easily pass through, the cell represents a 0; when in a low resistance state, it represents a 1.
One class of cross-point memory is phase-change memory (PCM), which has been in development for years.
In 2008, for example, Toshiba and SanDisk announced they were working on a "cross point" memory RAM chip. In 2013, the two companies demonstrated their cross point memory at the International Solid-State Circuits Conference.
In 2011, IBM announced it had produced PCM chips that could store 100 times the amount of data NAND flash could while also having 5 million erase-write cycles.
Even Intel and partner Numonyx announced their own PCM breakthrough in 2009.
In 2013, start-up Crossbar introduced a postage stamp-sized ReRAM prototype that it said could store 1TB of data; the company expected to go into large-scale production this year.Computerworld
So what's different about 3D XPoint? The companies said they invented "unique material compounds" and a unique cross-point architecture that is 10 times denser than conventional memory and is able to scale in ways PCM cannot.
Additionally, XPoint's memory cells are written to or read by varying the amount of voltage sent to each "selector," meaning if the voltage is high or low, it's a 1 or 0, respectively.
"This eliminates the need for transistors, increasing capacity and reducing cost," Micron said in its marketing material.
"The majority of resistive RAMs out there today are filamentary, so you have this statistical variation in how the filaments form each time you program it. On top of that it's difficult to scale and it's very slow," Meyers said.
Intel and Micron aren't yet talking about the materials they're using to create the new memory.
"The companies carefully skirted such questions. They were intentionally unclear about what it is," Handy said.
Gregory Wong, an analyst with Forward Insights, said Intel and Micron appear to be about two years ahead of other memory makers in the development of cross-point memory.
3D XPoint technology uses its new material to switch the resistance state, so it doesn't rely on less reliable and more expensive elements, such as Memristor's titanium dioxide and platinum films or PCM's silver filaments, which wear out over time. And, in that one difference, Intel and Micron said they were able to bring their product to the manufacturing phase.
"In addition, the combination of the architecture and the unique set of materials in both the memory cell and selector enable 3D XPoint technology to achieve increased density, along with significant improvements in performance and endurance," Micron wrote.
What it means for you and me
For consumer uses, XPoint could mean desktop and laptop storage would take up far less space with vastly more capacity, and it would boost today's NAND speeds from 500MB/s to 500GB/s if Intel's and Micron's numbers are to be believed.
Gartner Vice President Joseph Unsworth believes it will be years before consumers see 3D XPpoint in their PCs or laptops "because cost and performance is far more valuable to data centers and in-memory computing applications than it is for typical consumer usage, which is quite price sensitive."
Just as NAND flash once did, emerging memory technology can take years to catch on. The new memory, however, is bound to make a splash in data centers for applications such as in-memory computing and high-performance computing, Unsworth said.
Today, data centers use DRAM for high performance computing and other I/O intensive applications and capacitors to ensure that if power is lost, the DRAM modules have a few seconds of power to transfer the data to non-volatile memory. With XPoint memory, more high-performance processing could be done without the need for as much DRAM, eliminating the super capacitors and some NAND flash.
"I don't think you can replace DRAM entirely," Wong said. "What I think will happen is you'll still have a little bit of DRAM and then a huge amount of 3D XPoint memory."
3D XPoint could also indirectly benefit consumers. "The massive amount of data created and digested requires extensive analytics abilities and thus this could improve identifying fraud detection, transaction processing, genomics research, oil/gas mining exploration, IoT implications, etc," Unsworth wrote in an email reply to Computerworld.
All about that speed
Another important attribute is that 3D XPoint memory is fast, hyper fast.
Currently, memory and flash storage rely on two separate interfaces to a computer. A laptop or desktop typically use SATA bus interface or is connected directly to the motherboard via a PCI Express (PCIe) expansion slot. The DRAM or system memory then sits in its own board slots directly connected to the CPU. The point is that both have different interfaces and performance levels.
Intel and Micron are claiming that their 3D XPoint memory could usher in entirely new system architectures by replacing both DRAM and NAND flash.
For example, DRAM's latency falls into the nanosecond range, while NAND flash's falls in the microsecond range, which is still a thousand times faster than a hard drive's I/O speeds. While not offering specifics, Micron and Intel said 3D XPoint approaches the speed of DRAM.
The new material being used in 3D XPoint is critical. NAND memory has been approaching a lithography wall, meaning its transistors can't get much smaller. Currently, the smallest lithography process is between 10nm and 20nm in size. So NAND flash companies such as Intel, Micron, Samsung, SanDisk and Toshiba have been building 3D NAND, which stacks as many as 48 microscopic layers atop one another to increase memory density and capacity.
3D XPoint memory arrays are made up of perpendicular conductors that connect 128 billion densely packed memory cells. Each memory cell stores a single bit of data. The compact structure results in both the high performance and high density of the new memory.
Initially, XPoint memory will store 128Gbits per die across two stacked memory layers. "Future generations of this technology can increase the number of memory layers and/or use traditional lithographic pitch scaling to increase die capacity," Micron said in its marketing material.
"For decades, the industry has searched for ways to reduce the lag time between the processor and data to allow much faster analysis," Rob Crooke, general manager of Intel's Non-Volatile Memory Solutions Group, said in a statement. "This new class of non-volatile memory achieves this goal and brings game-changing performance to memory and storage solutions."