Tech by VICE

In New Advance, IBM Engineers Put Universal Memory Within Reach

Researchers solve key problems in the development of phase-change memory.

by Michael Byrne
May 19 2016, 11:00am

Image: IBM Research

It can be kind of hard to keep track of the tangle of emerging computer memory technologies with one seemingly announced every other week, from Intel's highly publicized but still mysterious 3D XPoint technology to organic ferroelectrics. The memory holy grail, however, is to reduce it all, the entire hierarchy of computer memory, to one single level. This is the universal memory dream.

Researchers at IBM announced this week at the IEEE International Memory Workshop in Paris—and in an accompanying paper—the successful storage of three bits of data per cell within a 64,000 cell phase-change memory (PCM) array. The bits were able to survive both high temperatures and up to one million endurance cycles, prerequisites for real-world utility.

To recap, computer memory is currently a bit of a mess. Or, rather, it has been since computer processor speeds started improving at a faster rate than memory access speeds. This disjoint has created what's known as the memory bottleneck, in which faster processors matter less and less as it becomes impossible to serve them data at corresponding speeds. It's sort of like piloting a racecar through rush-hour traffic.

The memory hierarchy begins at the actual computer processor(s), where just the tiniest amounts of data are stored very expensively in local registers representing stuff that is at that very moment being computed by the processor. Just above that are a few CPU memory caches, which are also very fast, very small, and very expensive (but a bit less so). Next is the fastest non-processor memory pool, which is RAM. This is the computer's main memory and it requires a continuous power input to maintain its current data stores.

Then we have some solid state drives and flash memory, which is faster than hard-drive memory and nonvolatile (data doesn't need refreshing by a power source to persist), but still part of the general pool of disc storage. At the bottom are regular old hard-drives.

Computer engineers would love for this whole pyramid to just collapse into one idealized form of memory that would be fast enough to serve as main/system memory, as inexpensive as disc storage, and as nonvolatile as disc storage. This is where things like PCM come in (PCM is just one of several universal memory-type possibilities being considered).

"Phase change memory is the first instantiation of a universal memory with properties of both DRAM and flash, thus answering one of the grand challenges of our industry," offers Haris Pozidis, manager of non-volatile memory research at IBM Research, in a statement. "Reaching 3 bits per cell is a significant milestone because at this density the cost of PCM will be significantly less than DRAM and closer to flash."

PCM is based on materials that change back and forth between crystalline and amorphous phases in response to an applied current. These different phases, which are analogous to melting and freezing, are used to encode data as bits, and, because there are several possible intermediate phases (between crystalline and amorphous), it's possible to encode several bits within the same cell. Information is then recovered by passing voltages through the memory cells, with different resistivity thresholds corresponding to different cell states. This is the same (basic) idea exploited in Blu-Ray disc technology.

The advances that enable IBM's PCM technology had less to do with the materials themselves than how information stored and recovered from the memory arrays is processed. On the one hand are improved metrics for determining a memory cell's current state, and on the other, are coding and detection schemes resistant to the material's natural drift between amorphous and crystalline states, which has the effect of changing the threshold voltages corresponding to the material's different information-encoding states.

PCM is still a work in progress, and by no means is IBM the only player. It's been assumed/inferred that Intel's XPoint is based on similar ideas, but said IBM competitor has been much less forthcoming about its technology.