Later this year, Sandisk plans to release flash devices with X3 and X4 technology which will allow 32GB MicroSD cards and the lowest-cost flash ever produced, although with potential lifespan and performance issues. Is X4 the new MLC?
By Ari Allyn-Feuer | Last updated February 10, 2009 1:00
Sandisk today announced it has developed several new flash memory devices based on its X3 and X4 technology, which stores three or four bits per NAND cell, up from the usual one (SLC) or two (MLC). The flash firm plans to commercialize its X4 technology next quarter and release new X3 devices later this year.
X3 and X4 were developed by an Israeli company called M-Systems, which licensed the technology to Samsung in 2003. The two firms entered a legal dispute on licensing terms, and shortly thereafter Sandisk bought M-Systems for $1.5 Billion. Sandisk inherited the legal battle with Samsung, which it won in private arbitration in October. Samsung's attempt to buy Sandisk, which Sandisk painted as an attempt to buy its valuable technology on the cheap, fell apart several weeks later.
This left Sandisk firmly in control of a body of technology it believes holds the key to the future development of flash memory, but in rather poor financial condition. This announcement represents Sandisk's attempt at ruling the market with their new technology, or at convincing its flash competitors to license the technology in what could be a very lucrative arrangement for Sandisk. The flash firm is trying to make a display of the technology's potential.
Sandisk will announce two new flash devices with X3 and X4 technology at the International Solid State Circuits Conference in San Francisco today. Ars got on the horn with Sandisk Senior Vice President for Memory Design Khandker Quader to discuss them.
Later this year, Sandisk will ship a 64Gb chip monolithic flash chip with four bits per cell. This will, Sandisk claims, be the first 64Gb chip ever commercialized. Although it is manufactured on the aging 43nm process of Toshiba, Sandisk's manufacturing partner, use of X4 technology allows the new module to achieve density and performance parity with its 32nm competitors on the older process at lower cost, an achievement of which Sandisk is justifiably proud.
In the second quarter, Sandisk will begin shipping a 32 Gigabit monolothic flash chip with three bits per cell. Manufactured on Toshiba's upcoming 32nm process, which is tied for smallest in the industry, and carrying three bits per cell, the chip will carry this high capacity on only 113 square millimeters of silicon, small enough to fit on a MicroSD card and smaller than any current 32Gb product. With eight dies stacked on top of each other in a single card less than one millimeter thick, this will allow Sandisk to ship the world's first 32GB MicroSD card, cost-comparable with current cards on a capacity basis. This represents a quadrupling of capacity over existing 8GB products and a doubling over the potential capacity of current technology. To all these advantages, though, there are some costs. With less voltage difference between adjacent levels, the intrinsic accuracy of reading the differences between them is dramatically reduced.
To combat the lower intrinsic accuracy of higher-level MLC, Sandisk has designed a new flash controller. Controllers designed for current flash won't work with X4 products; too many errors would occur. Sandisk has, Quader told Ars, designed a new controller which uses a proprietary new ECC algorithm and signal processing techniques to overcome X4's inherent accuracy weakness. Similar controller advances were necessary in the early days of MLC flash, which Sandisk would probably have us retronym X2.
This new controller and algorithms, Quader claims, can make themselves useful even on MLC and SLC flash, by allowing higher accuracy and longer lifespans through more rigorous error correction. If this turns out to be correct, even firms which don't adopt X3 and X4 may find Sandisk's controller technology interesting.
Performance is negatively impacted by X3 and X4, too, although Quader says the higher density achieved by X3 and X4 products offsets this. The X3 part announced for 2H09 has a 5.6MB/s write speed, while the X4 product due later this year will have a more compelling 7.8MB/s write speed. If these speeds scale linearly, a 32GB MicroSD based on the X3 part would sustain writing at 44 MB/s, on par with the fastest available SD cards. A 128GB SSD with the X4 part would sustain writing at 125MB/s, in league with current SSDs. If Sandisk's performance claims pan out, X3 and X4 could be a seamless transition, performance and lifespan congruent with MLC while offering much greater density at lower cost.
This is somewhat surprising, and there is reason to be a bit skeptical. The transition from SLC to MLC saw dramatic deficits in lifespan and performance, deficits which only this year began to be resolved enough to allow MLC flash to be used in SSDs for hard disk applications. Enterprise users are still wary of them, a problem with which SSD vendors like OCZ and Intel still struggle. To see the X3 and X4 transition occur so smoothly, which, to listen to Sandisk, it will, is quite remarkable. If so, it speaks well of the engineers at M-Systems and Sandisk who built these devices.
In October, I wrote that "If SanDisk retains control of its patents as an independent company, and if X4 is as important as it seems it may be, all the NAND manufacturers may be forced to pay billions of dollars to the now-beleaguered flash firm." Sandisk has broken free of Samsung's purchase offer and survived as an independent company with control of X3 and X4, and with these new implementations of its patents is taking its best shot at making this vision a reality.