Almost ten years on, flash is still maturing and could last even longer than 10 years, Huffman said. Its adoption will grow in enterprises and client devices, and it will ultimately overtake hard drives, which have peaked on speed, she said.
Like Huffman, observers agreed that flash is faster and more durable, but also more expensive than hard drives. But in enterprises, SSDs are inherently parallel, and better suited for server infrastructures that need better throughput. Multiple SSDs can exchange large loads of data easily much like memory, Huffman said. SSDs can be plugged into PCI-Express 3.0 slots in servers for processing of applications like analytics, which is faster than hard drives on the slower SATA interface.
The US$30 billion enterprise storage market is still built on spinning disks, and there is a tremendous opportunity for SSDs, said Neil Vachharajani , software architect at Pure Storage, in a speech.
Typically, dedicated pools of spinning disks are needed for applications, which could block performance improvements, Vachharajani said.
"Why not take SSDs and put them into storage arrays," Vachharajani said. "You can treat your storage as a single pool."
Beyond being an alternative primary storage, NAND could be plugged into memory slots as a slower form of RAM. Facebook replaced DRAM with flash memory in a server called McDipper, and is also using SSDs for long-term cold storage. Ultrabooks use SSDs as operating system cache, and servers use SSDs for temporary caching in servers before data is moved to hard drives for long-term storage.
Manufacturing enhancements are being made to make SSDs faster and smaller. Samsung this month announced faster V-NAND flash storage chips that are up to 10 times more durable than the current flash storage used in mobile devices. The flash memory employs a 3D chip structure in which storage modules are stacked vertically.
Intel is taking a different approach to scaling down NAND flash by implementing high-k metal gate to reduce leakage, according to Krishna Parat, a fellow at the chip maker's nonvolatile memory group. As flash scales down in size, Intel will move to 3D transistor structuring, much like it does in microprocessors today on the 22-nanometer process.
But there are disadvantages. With every process shrink, the endurance of flash may drop, so steps need to be taken to preserve durability. Options would be to minimize writes by changing algorithms and controllers, and also to use compression, de-duplication and hardware encryption, attendees said.