SAN JOSE, Calif. – Samsung announced at the Flash Memory Summit chips, solid-state drives (SSDs) and systems designs geared to drive 3-D NAND into mass markets. But analysts and even other vendors said it could be another year or more before the technology is ready for the mainstream.
This month Samsung will start making a version of its 850 EVO SSDs using its latest 48-layer, 3-bit per cell, 256 Gbit 3-D NAND chips, said Jim Elliot, a marketing executive for Samsung, speaking in a keynote here. The new chips sport twice the performance and more than 50% better power efficiency than the company’s current 32-layer chips, he said.
Samsung believes in can make future Tbit-class chips with more than 100 layers, he said. In a separate talk, an SK Hynix engineer said the company will start production in the third quarter of 3-D NAND chips that sport more than 30 layers and by 2019 make chips with more than 190 layers.
The dense chips promise many generations of continuing improvements in flash performance and decreases in costs. However today 3-D NAND yields are still low and costs are still higher than traditional planar flash chips, said analysts.
Although Samsung was the first to announce production of 3-D NAND, its rivals are catching up. Last week Toshiba announced it will sample in September a 256 Gbit 3-D NAND chip with three bits per cell and 48 layers.
"People are making [3-D NAND] announcements faster than I can put them on a slide," said analyst Mark Webb pointing to this compilation.
Samsung is shipping its V-NAND SSDs at a loss, and most vendors say a big ramp for 3-D NAND won’t come until 2017, said Jim Handy, an analyst at Objective Analysis (Los Gatos, Calif.), speaking in a session here. The 48-layer devices, once they enter volume production, should be the first to beat traditional flash on cost, said Mark K. Webb, a consultant at MKW Ventures (Albuquerque).
The 32-layer 3-D chips have 70% higher wafer costs than traditional flash, but the 48-layer versions get 50% more layers for only 5-10% higher wafer costs, Webb said. The chips tend to start off at 50% yields but after a year or so approach planar flash yield levels, he said.
Webb predicted 3-D NAND won’t make up the majority of flash bit sales until 2018 or later. “It takes a couple years for a new technology to mature,” said Jetming Woo, a senior manager for SSD marketing at SK Hynix in a talk here.
Vendors are still experimenting to find what number of 3-D layers will provide optimal products, Woo said. In addition, they have to develop a new class of controllers and firmware to manage the larger block sizes and other unique characteristics of 3-D chips, he added.
Next page: V-NAND powers fast, dense drives