Another point of view with some additional colors:
--------------
THE NEW COMPUTER memory from Intel and Micron is deceptively simple.
Typically, memory chips store information in transistors, tiny three-pronged devices that can house electrons. But with their new memory tech, unveiled on Tuesday, Intel and Micron have done away the transistor, storing information in what is really just a lattice of wires. Each piece of information sits where two wires cross.
“This is the simplest structure you can build,” says Al Fazio, the Intel Fellow who serves as the director of memory technology development at the chip giant. “It’s just the intersection of two orthogonal lines.” The companies call it 3D XPoint—pronounced “3D cross point”—because those intersecting wires look like Xs.
The trick is that after years of research—dedicated work on the particular project began in 2012—Intel and Micron say they can now build these chips from materials that can store and retrieve unusually large amounts of data at unusually fast speeds. According to the two companies, the chips are about 1,000 times faster than the flash memory that underpins your iPhone and can store about 10 times more data than the DRAM memory in PCs, laptops, and servers.
Chip companies tout new memory tech all the time. HP has been banging the drum for its “memristor” technology for years—with little real product to show for it—and Phillip LoPresti, the CEO of a company called Everspin, tells us his company is already offering a memory technology behaves a lot like 3D XPoint. But Intel and Micron are claiming a unique blend of memory characteristics, and there is a very real need for the kind of chip they describe. The biggest customers may be internet giants like Google, Facebook, and Amazon.
The Third Dimension
Yes, 3D XPoint is terrible name. But that’s the way things are done in the tech world. And at least this name is descriptive. The “3D” bit indicates that those cross-point wire lattices can be stacked on top of each other (see image above).
Basically, these chips store data by changing the electrical resistance at those many cross points. This is done with a tiny contraption that Fazio calls a “selector.” This is kinda like a diode, which can switch from low resistance to high.
Screen Shot 2015-07-29 at 10.13.13 AMClick to Open Overlay Gallery
INTEL
That’s quite different from memory technologies like DRAM and Flash, which use transistors. A transistor is essentially a switch made from three basic components: a source, gate, and drain. When a certain voltage is applied to the gate, current flows from the source to the drain, and the transistor is “on.” Apply another voltage, the current stops, and the transistor is “off.” 3D XPoint doesn’t move current like this. “It changes the property of the material,” says Cooke. “It doesn’t try to store an electron.”
This simple architecture, Fazio says, is what allows the chips to store so much more data than DRAM (per area). They still aren’t as fast as DRAM, but unlike DRAM, they’re “non-volatile,” meaning they can retain data even when a machine is powered down. Flash is non-volatile too, but 3D XPoint is significantly faster than flash. That extra speed, Fazio says, comes from the unique combination of materials that Intel and Micron use to build the lattice and the selector.
A Real Need For Speed
Intel says 3D XPoint will be great for gaming machines—“today, game designers are restricted in what they can do by the capabilities of the system,” says Rob Crooke, Intel senior vice president and general manager of the company’s non-volatile memory solutions group—and that may well be the case. But the real need is inside the massive computer data centers that power the most popular web services, including Google, Facebook, and Amazon.
In the data center, these companies must spread information across thousands of machines. In an effort to speed the storage and retrieval of this data, they’re replacing traditional hard drives with faster flash memory, and in some cases, they’re running databases that keep data in DRAM, sidestepping both hard drives and flash. As described by Intel and Micron, 3X Point can provide still more performance inside these warehoused-sized computing centers—something the Facebooks and the Googles are always hungry for.
“They do a lot of in-memory, and they do it across many systems,” says Patrick Moorhead, president and principal analyst at the chip-centric research firm Moor Insights and Strategy.
At the same time, Moorhead says, the new tech can help more traditional businesses. The typical enterprise doesn’t store data across thousands of machines the way a Facebook or a Google does. But smaller outfits are running “in-memory databases”—i.e. in-DRAM databases—on individual machines or small numbers of machines. 3D X Point can potentially help here as well. It can provide more memory per machine.
The Materials Secret
Will this new technology work advertised? We’ll have to wait and see. Intel and Micron say that first machines equipped with 3D XPoint won’t arrive until next year, and the two companies won’t even say what materials they’re using to build these chips.
But that’s telling. Intel and Micron don’t want anyone else imitating what they’ve done. Remember: this is a rather simple architecture. The companies intend to both manufacturer and sell the chips on their own. They jointly run the Utah plant where the chips will be built.
That means they think they’re onto something. But as Moorhead points out, this also indicates that this will very much be a tech for the world of data centers. “IT is their business,” he says of Intel and Micron. Data centers are complex things. But sometimes, simple technologies can improve them in big ways.
AI
