Flash Storage Innovations Leading to Lower Prices

These three innovations in solid state storage technology mean flash prices will continue to drop for the foreseeable future.

High-performance flash-based storage is increasingly insinuating itself into enterprise infrastructure. It's showing up in everything from server caches and directly attached storage to hybrid arrays and fully solid state storage appliances.

But corporate storage requirements are exploding, meaning enterprises will need far more storage capacity — both flash and spinning hard disk storage — in the future.

Hard disk drive (HDD) capacities continue to grow, while the decrease in the price per gigabyte shows no sign of slowing, thanks to numerous innovations ranging from new ways to write data like shingled magnetic recording (SMR) and heat-assisted magnetic recording (HAMR) to the practice of filling drive cases with helium.

The big question is whether a similar range of innovations are appearing in the world of solid state storage. Can solid state drive (SSD) manufacturers increase the capacity (and reduce the cost) of their products at a similar rate to HDD makers, so that solid state storage can still make economic sense in the large quantities that will be required in the future?

The good news is that three important technology innovations will allow flash storage to keep up with spinning disks in terms of capacity increases and falling cost per gigabyte.

Storage TLC

The first of these innovations revolves around cell levels and relates to the way that flash memory stores data.

Flash storage is made up of memory cells, each of which has a floating gate transistor that can hold a charge. The state of this transistor determines the data that the cell is storing. Currently there are a variety of different types of flash, including single-level cell (SLC), multi-level cell (MLC) and enterprise grade MLC (eMLC). Understanding these is the key to understanding future innovation.

SLC is the simplest flash technology, and it involves cells that each hold a single bit of data. That's because the cell's transistor can hold one of two possible charges, corresponding to either a 0 or a 1. The primary benefits of SLC flash are that it is hard wearing — meaning that many read/write cycles can be carried out before the cell is worn out and no longer works — and that reads and writes are fast. The drawback is that it is expensive as only a single bit of data can be stored in each cell. Because it is expensive, SLC flash is often called enterprise grade flash, and it used in premium storage products rather than cheaper "consumer grade" devices.

MLC is more complex than SLC because each cell's transistor can hold four different levels of charge, allowing two bits to be stored in the cell. (The four charge levels correspond to both bits being 0, both being 1, 1-0 or 0-1.) MLC tends to be at least half the price of SLC (because more data can be stored per cell) but it wears out an order of magnitude quicker. And write speeds to MLC are slower than to SLC. For those reasons MLC tends to be used in consumer devices rather than expensive enterprise-grade hardware.

Because MLC is so much cheaper than SLC, manufacturers came up with the concept of eMLC — multi-level cell flash that was designed to be (almost) as hard wearing as SLC, making it more suitable for enterprise environments.

According to Jim Handy, a solid state storage expert and semiconductor analyst at Objective Analysis, the main difference between eMLC and standard MLC is just that the firmware in the hardware controller that manages the flash cells writes to the cells more "gently." This results in around three times the endurance of MLC, but at a cost of 15 to 25 percent lower performance, he says.

The latest innovation in cell levels involves taking MLC a stage further with the development of the triple-level cell (TLC). As the name suggests, this technology has the ability to cram three bits of data in each cell, using a transistor that can store eight different charge levels. Because it can store three times the data in the same number of cells, TLC is very much cheaper per gigabyte than SLC, and it also enables more storage capacity to be built into a given form factor.

There are, however, drawbacks to TLC flash, and once again these are related to speed and durability. But what's exciting about TLC is that innovations in flash controller technology are making these drawbacks far less important.

Controller Tech

A flash controller chip manages the data stored on flash memory and communicates with connected devices. Current flash controller technology is improving very rapidly. To a large extent these improvements are possible because falling costs have made it practical to use increasingly powerful controller chips that can run more sophisticated firmware faster.

That means that eMLC is no longer really necessary, according to Handy. "Most enterprise SSDs now use standard MLC rather than eMLC or SLC, because the new controllers have erased the advantages that these other technologies offered. With a better controller you get the same write speeds and better wear," he says.

In other words, improvements in controller technology more than offset the inferior speed and durability characteristics of MLC. "That means that today's SSDs are more reliable and have better wear than ever. Today's MLC is better than yesterday's SLC," he explains.

This progress is continuing, which has important implications for overcoming the main drawback of triple-cell technology: TLC is more prone to errors and thus needs much better error correction technology than was the case with MLC or SLC.

Such error correction technology is available, in the form of a powerful but computationally complex error correction algorithm called Low Density Parity Check (LDPC). (LDPC was originally invented in the 1960s, but it required more processing power than was available at the time, so it was largely ignored then.)

"LDPC requires a far higher gate count in the controller hardware, so implementing LDPC may add 30 percent to the cost of the controller," says Handy. But thanks to falling controller hardware prices, it is now becoming practical to implement LDPC in controller silicon, and some manufacturers are now starting to use it in their controllers.

As a result of this type of hardware innovation, TLC is becoming increasingly attractive, and it is likely to make up 50 percent of total flash production by the end of 2015, according to research from DRAMeXchange. "I do believe that eventually everything will be using TLC," says Handy.

Third Dimension

The other major flash innovation on the horizon is the development of 3D flash: memory that is built in layers stacked on top of each other with vertically-oriented cells. Current technology involves 2D planar flash, which is made in a single layer with cells that are oriented horizontally.

"3D flash is on all the manufacturers' roadmaps. That's because they have realized that they can no longer increase capacity by shrinking the size of the cells in the way that they have been doing," says Handy.

Because 3D flash allows for many more cells in a given surface area (because they are aligned vertically and stacked on top of each other), overall capacity can be increased while actually using a larger cell size. Typical planar flash cells are made using a 15nm process, while 3D cells are currently made using a 45nm process. This larger size has the added benefit of making 3D flash much less subject to errors due to interference from neighbors and therefore more reliable than smaller cells. Samsung has already started producing SSDs that use 32-layer 3D TLC to produce 1TB drives, and in March Toshiba announced that it is developing 48-layer 3D flash using MLC. Improved fabrication techniques are likely to enable manufacturers to make 3D flash (using MLC and TLC) with even more layers in the near future.

"We don't know what the limit to the number of layers is yet, but today it's thought that it is about a hundred," says Handy.

From an economic point of view it's interesting to note that the prices per gigabyte of both HDDs and SSDs have been falling at about the same rate for at least the last seven years. Flash storage has remained about twenty times more expensive per gigabyte than spinning disk storage during that time, Handy says.

The good news for enterprises with rapidly expanding storage requirements is that improvements in cell technology, enhanced controllers and 3D flash mean that flash storage costs are set to continue falling rapidly in line with HDDs for the foreseeable future.

Photo courtesy of Shutterstock.