Network Computing is part of the Informa Tech Division of Informa PLC

This site is operated by a business or businesses owned by Informa PLC and all copyright resides with them. Informa PLC's registered office is 5 Howick Place, London SW1P 1WG. Registered in England and Wales. Number 8860726.

What's Next For Flash?

Flash is just seven years old this year, and applications are just beginning to really take advantage of it, so a discussion of its demise and the arrival of alternatives may seem premature. The reality is that flash opened Pandora's Box on the storage front and upset a long period of stagnation. Innovation is happening fast now in the storage industry.

In addition to efforts to evolve from 2D flash to 3D flash, the industry is developing at least seven storage technologies touted as faster and denser successors to today's flash.

The developments come as flash is starting to hit the limits of physics. Both the density and the speed of current technologies are topped out. The density issue is leading to a generation of 3D NAND where storage cells are stacked vertically on a die, instead of horizontally. This will increase capacity per die into the terabit range, though it adds cost, both in processing and in flaw management.

Stacking die is another option that's attractive. The driving forces are performance and power dissipation. Using a technology called "through-silicon via," die can be stacked on top of one another with connections going from one layer to the next. This technology becomes really interesting with the Hybrid Memory Cube (HMC) architecture, which uses many serial links for the through-via communications and stacks DRAM or flash together.

HMC uses very little power, saving perhaps 80% over today's solutions for DRAM. It's also much faster. The current specification envisions terabyte-per-second speeds.

With 3D flash, the main result will be small devices with lots of memory, perhaps including 1-inch SSD and multi-terabyte 2.5-inch products. Throw in HMC, and we will see memory stacked on to CPUs directly, with capacities in the terabyte range for DRAM and multiple terabytes for flash.

Super-chip modules such as these will power in-memory databases and HPC systems, and they will expand virtualization and hosted desktops, but there will be spinoffs once the TSV process gets ironed out and is cheap enough to be mainstream. These include baby super-chip stacks for mobile devices, for instance.

For all these gains, however, flash is less than ideal as a DRAM extender. It's just too slow. Writing is often sped up in flash or SSD devices by using some DRAM as a write buffer, but this doesn't handle the read speed mismatch that we would see in HMC, and even the best caching algorithms can't solve that problem.

This is where the next innovations in solid-state storage become important. Most of these emerging technologies boast much faster speed, getting close to DRAM levels. Most are denser, at least in theory. But most are still in the research labs.

The likeliest contenders are spintronic memory, which is in production in very small capacities, and resistive RAM. There are claims for demonstrations of terabit spin memory, but the money is on ReRAM winning the race, at least at the moment. With that technology, we might get to half the latency of a DRAM cell. Combine it with the parallel access approach of HMC, and that would be very acceptable performance.

In the longer term -- perhaps a decade out -- graphene interconnect and its use as a substrate hold promise for speeding up CPU and memory transistors while drastically reducing power. This would allow 3D stacking of CPU cores and DRAM pages. It would also open up a market for very large, inexpensive solid-state drives as replacements for today's spinning disk bulk storage.

Flash underpins much of the change occurring in IT today. That's why it sees such strong evolutionary pressures. Persistent solid-state memory is clearly a technology area to watch, and how it moves forward will affect the whole IT community in major ways.