Intel is introducing a new type of memory component. They refer to it as a "disk" although it actually more like a flash drive. The technology behind this component is still (slightly) veiled behind corporate secrecy, but it seems to be based on a phase change material. It is similar to be technology I wrote about several months ago (5D Crystal Technology: Huge Data Stores That Will Last Till the End of Time ) which is based on glass-like substances that alter their physical atomic arrangement when heated.
Intel is calling this new technology Optane, and is based on some joint development they have done with Micron (the memory manufacturer). The underlying joint technology is called 3-D Xpoint, and while they haven't publicized exactly how it works you might want to check out A Preview of Future Disk Drives which gives a clearly written, high-level overview of a working research prototype (non-Intel) based on phase change principles.
Benchmarks show that the technology can provide higher data density by volume as well as a 10 times speed of NAND memory. In the optical data channel becomes necessary in order to keep up with the data transfer rates that the 3-D Xpoint chip is capable of. And, even more interesting, Intel promises that in 2017 that they will repackage this technology into standard RAM memory slot modules!
One reason I find new technologies like these fascinating is because they represent a continuum of storage solutions. In the very earliest days computer storage began as permanent and very slow (punched cards), or volatile but much faster (electrical relays). The slow storage technology got bigger and faster: paper tape, magnetic tape, block access magnetic tape, disk drives, solid-state drives. And this "slow" style of storage has promising new avenues such as the 5D Crystal I mentioned previously. And let's not forget DNA! The faster volatile storage paradigm has also gained capacity and speed. Electrical relays soon led to vacuum tube digital switches, transistor switches, integrated circuit arrays of transistor switches (static RAM), capacitor based transistor arrays (DRAM), memristors.
But as these memories solutions evolved they often had some interesting and quixotic crossover too. At the time of discrete transistor switch memory and block access magnetic tape technology there was a need for low latency (less than 10 ms) random access to larger (1000 bits!) blocks of data. This led to the now-defunct class of components called magnetostrictive delay line memories. They were sort of like a disk but instead of spinning discs a wave of magnetic fields was propagated around a circle of wire (usually nickel alloy). When the pulses (representing bits) reached the other end of the wire they were detected, amplified and reinjected at the starting point. Knowing the clock counter it was possible to read any desired bit out of the rapidly cycling train of bits. It behaved like a disk but it was volatile. If you lost power you lost all the data.
In case you're wondering what on earth could be done with such limited and primitive technology, magnetostrictive delay loop storage was the core of the most advanced numerical control milling machine of its time (the late 1960s). It was used to mill the giant full span spars for the first Boeing 747's.
Another reason these new storage technologies are interesting is because they all are tending towards long-duration, non-volatility. Some of you probably noticed that when I listed the spectrum of fast random-access storage I left out an entire class of memory that was successful and cost-effective from the late 1950s through the 70s: ferrite core memory. Until the advent of flash memory, it was unique in that it would retain data indefinitely with no power. But issues with manufacturing, quality control, grid sizes, etc. imposed definite limits on the scalability of this type of memory. Once IC based memories became available they quickly supplanted core memories even though the IC memories were volatile. Interesting aside: The first space shuttle flight computers use core memory and after the Challenger crash useful data was recovered after being brought up from the ocean floor.
While all of these technologies are discrete it becomes apparent (once we have enough of them) that there is a continuum across the speed of access and quantity of storage spectrum. And as we get more nonvolatile, low cost, random-access, low-power, large capacity storage technology the sooner the distinction between cheap disks and fast RAM will disappear. It will definitely change the way we write programs.
Maybe in 2017 and with some Optane chips I can start doing 1 TB matrix transformations in memory on my desktop PC!