Flash Memory-The Great Disruptor – ISSCC2012
February 20, 2012, ISSCC, San Francisco—Eli Harari, the founder of SanDisk described flash memory as a disruptive force in the industry.
Nonvolatile memory has been around for a really long time. The first examples were clay tablets which required high effort for one-time programming. More recently, in 1970 Intel invented the EPROM which was programmable with great effort more than once, but erasure required exposure to high intensity UV light. EPROMs (erasable programmable read-only memory) enabled the microprocessor revolution.
In 76, Hughes invented the EEPROM, and electronically erasable memory. These devices were based on thin oxides and retained data for a long time based on their floating gate architecture. A two-year evaluation efforts revealed the physics associated with the high effort programming any race cycles. The thin oxide film eventually breaks down due to electron trapping.
In 84, Exel and Intel introduced flash EEPROM configured as NOR flash. These devices used hot electron tunneling for writing and erasing. They were capable of fast reads but were very slow for write and erase, so they proved to be unsuitable for data storage. Seven years after, in 87, Toshiba introduced NAND flash, which at the time was a solution looking for problem. Eventually this technology became the primary medium for portable data storage.
In 88, SanDisk introduced system flash as HDD emulation. These devices featured fast writes and erases, serial read and low error rates. Data retention was measured in decades and the design allowed for low-cost storage solutions. The rapid process scaling in the semiconductor industry help to reduce costs and raise capacities over very short periods of time. To continue increasing capacity at the same rate now requires the flash to store multi-levels in a single cell.
In a system configuration, these new memory systems required new controllers to dynamically manage defects and wear-out leveling, and other aspects of these new devices. The rationale for going to the NAND structure over NOR, was that NOR flash is considered to be unreliable, and this new type of controller could compensate for the reliability and management issues associated with NAND.
In 91, they introduced the first SSD, a 20 MB 2-1/2 inch ATA storage device. Based on the physical size and storage capacity, they achieved about 3 MB per cubic inch. The latest product is now capable of holding 64 GB in a micro SD format and achieves about 6 TB per cubic inch. In 99, SanDisk and Toshiba formed a partnership to pursue ongoing NAND developments. Process scaling and new storage architectures have driven capacities from 256 MB to 64 GB and, soon to be announced, 128 GB.
Looking forward into the future, semiconductor processing will continue at least to the 10 nm node and possibly beyond that. The device people are continuously working on new cells and internal architectures to take advantage of the scaled semiconductor processes, and also to address reliability, wear out, and retention issues. One challenge that is not insignificant, is continued process scaling will require EUV for feature imaging. A new fab to process 10 nanometer features will cost more than $10 billion.
Another interesting technology for nonvolatile memories is resistive RAM. This technology has the potential to displace both flash and hard disk drives. Manufacturing storage using 3-D resistive RAM could provide capacities, performance, and costs comparable to HDD's. At the other end, the lack of wear-out mechanisms and the very high speed write and erase capabilities competes well with flash.
To date, and looking into the future, nonvolatile memories have enabled iconic products. The first large-scale product to use flash was the iPod where flash replaced a 1.8 inch HDD. The cost of flash has dropped by a factor of 50,000 over last 20 years and will continue to generate waves of disruption. Over time, storage media change and always move to the most cost-effective technology available.
Market drivers for new generations of nonvolatile memory technologies are digital consumer products, the Internet, and mobile devices. This is a very price elastic market and covers about 2 1/2 billion consumers. The evolution of industry standards such as SD, eMMC, and USB flash drives have resulted in host and flash independence through integrated controllers.
These embedded controllers allow any platform to host the storage, and the host device doesn't care about the nature of the flash storage. Mobile devices are becoming the new personal computer and at the same time are the ultimate convergence device. The industry needs flash for functionality in future devices.
SSDs change the storage architecture in the PC and have enabled all the mobile devices. SSDs are moving into the enterprise and into the cloud as I/O devices and as power reduction alternatives to other drive technologies. Notebooks and now the new Ultra books use SSDs for fast on operations, and some requiring larger storage capacities are using hybrid SSD and HDDs.
Flash is used as cache in the cloud and also can be used to timeshift and synchronize data transfers to bandwidth limited mobile devices. NAND will eventually overtake DRAM because of its very large cost advantages. However, write time is still an issue, so hybrid DRAM flash systems will become more commonplace over time. By 2020, flash will be the king of storage and will eclipse both DRAM and HDD in total volume. Innovation leads to disruption which leads to more innovation.