Silicon Week: What's in store: The tech that will transform memory and storage

Introduction and 3D XPoint

A few months ago, we took an aging laptop and brought it kicking and screaming into 2016. We quadrupled the RAM, swapped the hard disk for an SSD and transformed it from a bit of a laggard into the electronic equivalent of Usain Bolt. Usain Volt, anyone?

Bad puns aside, you've probably experienced similar performance boosts as PCs and Macs boast ever more RAM and ever faster SSDs. But what's on the horizon is even more exciting…

The tech trinity

The performance of any device is largely due to three things: the CPU/GPU, the memory and the storage. While most of the attention is often focused on the processor and graphics solution, key developments are happening in memory and storage as well which will affect CPU and GPU tech, too.

One of the most promising new technologies is High Bandwidth Memory or HBM for short. Although it's a very new technology Samsung and Hynix are already developing the third generation, which they expect to commercialise in 2019 or 2020.

Unlike traditional memory, where chips are laid flat on the memory module, HBM chips are stacked. That shortens the distance between the chips and the CPU or GPU, achieving the same speeds as on-chip integrated RAM, and it enables manufacturers to cram more RAM into smaller spaces. And we don't just mean slightly smaller.

AMD, which originally created HBM and the firm's Fiji processors are the first CPUs to use HBM, reckons HBM takes up 94% less space than the equivalent GDDR5. Where 1GB of GDDR5 takes up 28 x 24mm of surface space, 1GB of HBM needs just 7 x 5mm. That's particularly exciting for virtual reality, as it means powerful GPUs could live inside the headsets without them being so heavy you can barely move your head.

As AMD explains: "GDDR5 has served the industry very well these past seven years, but as graphics chips grow faster, their appetite for fast delivery of information continues to increase." GDDR5 is good, but its ability to meet those requirements "is beginning to wane as the technology reaches the limits of its specification." HBM resets the clock, offering more than three times the bandwidth per watt of GDDR5.

Current HBM supports up to 8 dies per stack, 8GB per package and achieves speeds of 256GB/s. Third generation HBM will boost those numbers even further, doubling the density to 16GB per die and enabling stacks more than 8 times higher. The only negative is that you can't get it yet.

Heart of glass

According to Intel, 3D XPoint technology is "the first new memory category in more than 25 years." Up to 1,000 times faster than NAND flash storage, up to 1,000 times more reliable and capable of storing up to 128GB per die, it's the technology behind Intel's exciting Optane SSDs.

The details of 3D XPoint – also known as QuantX, which is the name preferred by Intel's partner, Micron – haven't been fully disclosed, although Intel does say that it is "not based on electrons"; it's believed to be based on 'phase change' technology that writes data by heating a glass-like material.

Phase change chips can write or rewrite individual bits, something flash storage can't do, and just adding a buffer of phase change material to existing flash storage technologies can deliver significant performance improvements. Intel promises to launch its Optane SSDs by the end of 2016, and to have 3D XPoint RAM for PCs next year.

Top Image Credit: Wikipedia (AMD's Fiji)

Spintronics and HAMR, SMR

Spin on this

Spintronics is a new, ahem, spin on data transfer. As electrons spin they generate tiny magnetic fields, and those fields can be used to transfer data. That data transfer requires much less energy than traditional electronic circuitry, it can be used in cheap materials such as copper and aluminium, and it's non-volatile, so it retains data when there's no energy source.

Spintronic magnetic RAM – MRAM – already exists. One manufacturer, Everspin, markets MRAM chips for automotive and aeronautical applications, which benefit from MRAM's exceptional resistance to heat and other stresses. It's ideal for flash storage such as SD cards and SSDs, and while it's a little slower than conventional DRAM it's much smaller and more energy efficient. Those benefits mean it can also be used as processor cache memory, which is something Intel, Qualcomm, Samsung and Toshiba are all experimenting with.

Life's a gas

Don't write off the humble hard disk just yet, though. Earlier this year Seagate unveiled a 10TB helium-filled hard drive, following in the footsteps of Western Digital subsidiary HGST. Helium is considerably less dense than normal air, so using it inside drives reduces drag and other forces. That means manufacturers can use thinner platters and pack more platters into the same space. According to Seagate, the mean time between failures (MTBF) of helium drives is 2.5 million hours, compared to 2 million for non-helium enterprise drives.

For all its joys, helium doesn't make hard disk platters store any more information. For that, we need to consider technologies such as heat-assisted magnetic recording, or HAMR for short. HAMR adds a laser to the familiar hard disk design, heating up the platter to cram much more data into the same space. TDK promised to ship HAMR drives in 2016, but the timescale appears to have slipped and HAMR drives aren't expected to turn up much before 2018.

HAMR has a rival: SMR, or shingled magnetic recording. Once again SMR promises more dense data storage on hard disk platters, but it accomplishes this in a different way. Think of the shingles you see on a roof. SMR does much the same with data, creating tracks that overlap part of the previously written track, and Seagate has been shipping drives that use the technology for the last three years.

It works by exploiting the difference between a hard disk's read and write heads: read heads are narrower than write heads, so the overlapping doesn't prevent the reader from getting the full information. Because it's very similar to existing hard disk technology the costs of making it are relatively low, and it enables firms such as Seagate to increase storage densities by around 25%. That makes it a useful step forward while we wait for technologies such as HAMR.

• Also check out: 10 CPUs that changed computing

August 26, 2016 at 02:30AM
Gary Marshall