SanDisk co-founder: Flash to squeeze out hard drives and DRAM by 2020

The co-founder of SanDisk and one of the illustrious fathers of flash memory, Eli Harari, says that flash memory will “checkmate” hard drives by 2020. This is in stark contrast to Microsoft Research and UCSD, which earlier this week claimed thatsolid-state storage would meet its maker by 2024.

Speaking at the International Solid-State Circuits Conference (ISSCC) on Monday,Harari not only proclaimed that NAND flash would supplant spinning-platter hard drives, but also that DRAM could be on the way out as well. “Today, the cost of NAND per gigabyte is 10 times lower than the cost of DRAM … and that’s not likely to change,” Harari said. “The question is, can 10 gigabytes of NAND or one gigabyte of DRAM give you a better performance boost?”

Flash has been the dominant storage medium for years in the mobile space — cheap NAND was one of the most important factors in the explosion of digital photography and smartphones — and through tablets, ultrabooks, and enterprise applications, SSDs are really starting to dig into the HDD market share. With the steadily declining price of solid-state drives and their far superior performance, it’s really no big surprise.

According to Harari, though, it will be 3D resistive RAM (3D-ReRAM) that results in “checkmate for the hard disk drive industry.” ReRAM is a very old tech, but for various reasons never made it to the limelight — until 2008, when HP created the first memresistor. In much the same way that Intel has moved to FinFET to scale beyond 22nm, 3D-ReRAM is expected to take over from NAND flash at around 11nm, sometime in the next few years.

It is anticipated that 3D-ReRAM will be so fast and high-density that hard drives will be reduced to specific use cases, much like magnetic tape. “I believe that by 2020, flash – -which is highly scaled NAND and 3D resistive RAM –- will be the undisputed king of storage,” Harari predicts.

Finally, Herari also notes that the emergence and increasing reliance on cloud computing and storage could pose an issue, especially for mobile devices. Ultimately, irrespective of how much flash storage we have, mobile bandwidth is finite. There’s no point having hundreds of gigabytes of ultra-fast flash storage both in the cloud and on your phone when it can cost tens of dollars to transfer a single gigabyte of cellular data over a few-megabit connection.

Source

Caltech sets 186Gbps internet speed record

A delicious combination of high-energy physicists, engineers, and computer scientists from Caltech and the University of Victoria have broken the world record for sustained, computer-to-computer transfer over a network. Between the SuperComputing 2011 (SC11) convention in Seattle and the University of Victoria Computer Centre, Canada — a distance of 134 miles (217km) — a transfer rate of 186 gigabits per second was achieved over a 100Gbps bidirectional fiber optic link; 98Gbps in one direction, 88Gbps in the other.

At first blush, when you remember that scientists have transferred terabits per second over fiber optics, you might question the veracity of this world record. In those cases, though, the transfers were done over private networks under laboratory/testbed conditions — Caltech and the University of Victoria set their virtual land speed record over a standard, commercially-available 100Gbps link. Furthermore, the scientists didn’t simply shoot data down the pipe and into the digital ether, which is relatively easy — there were computers on either end of the link!

At SC11, Caltech employed 13 computer servers, equipped with a variety of PCIe 2.0 and 3.0 solid-state drives and 40Gbps LAN connections; on the Canadian side, University of Victoria used 10 Dell servers each with 10Gbps Ethernet links and six SSDs. At its peak, the team achieved a disk to disk transfer rate of 60Gbps — around 7.5 gigabytes per second. The 186Gbps record (23.25 gigabytes per second) was a memory-to-memory transfer between the servers. The max per-computer speed was 35Gbps, performed by one of the PCIe 3.0 servers. The 60 and 186Gbps transfers, incidentally, were sustained for no less than 11 hours. For a complete listing of the hardware and infrastructure used, hit up the Caltech SuperComputing 2011 site. There’s an overly-dramatic video demonstrating the link-up at the bottom of this story.

By this point you are probably wondering what all of those gigabits are actually for. Seattleites could use the connection as the mother of all cross-border digital smuggling routes — you could transfer 100,000 Blu-ray rips per day — or, more realistically, it could be used by scientists and physicists to share vast amounts of raw data, like that collected by the CMS and ATLAS detectors at CERN’s Large Hadron Collider. So far, hundreds of petabytes of data has been created by the LHC, but the only way for scientists to get a hold of it is by downloading it across the web (or loading up a van with terabyte tapes and driving it across the world). Currently, the fastest links are via Internet2 — but even that can’t sustain more than 10Gbps.

The fact is, many large-scale science experiments produce more data than a single institution can handle — and the only way to get around that is by sharing the load. The link demonstrated by Caltech and the University of California, beyond a promising glimpse of future at-home internet connections, directly improves our chance of discovering awesome things like the Higgs boson.

Read more at Caltech

[Image credit: Chrys Omori]

Source