In terms of technology, manufacturers are pushing the boundaries, increasing memory densities in 25nm and even smaller process technologies. Major NAND flash memory manufacturers are beginning production of 64Gbit memories in technologies ranging from 20 to 30nm and are increasingly using innovative memory architectures and structures to meet high-density demands. A recent paper by SanDisk and Toshiba, presented at the ISSCC (International Solid State Circuits Conference) earlier this year, detailed a 128Gbit NAND flash device in a 19nm, 3-bit-per-cell technology.
Replacement Possibilities:
While flash continues to scale in the short and medium term, there remains ongoing demand for potential long-term replacements in both standalone and embedded applications. The competition includes up to 30 different non-volatile memory technologies being examined by semiconductor companies, research institutions, and universities. Some of these technologies have already reached relatively small sales volumes. Four of the leading technologies that offer several advantages over flash, such as read/write times 100 times faster and significantly greater write cycle endurance, are: Phase-Change Memory (PRAM); Ferroelectric RAM (FRAM); Magnetoresistive RAM (MRAM); and Resistive RAM (RRAM).
PRAM or PCM:
PRAM memory takes advantage of the behavior of chalcogenide glass, which allows switching between four different states: crystalline, amorphous, and two partially crystalline states. This switching is made possible by the heat generated: when an electric current passes through the glass, the material's state changes. Since PRAM memory can produce up to four states, it allows for doubling the storage capacity, providing two bits per cell. The different states have very different electrical resistivities, so, for example, the high-resistance amorphous state can represent a "0" and the low-resistance crystalline state a "1".
The main disadvantage of PRAM is its sensitivity to high temperatures. It also degrades with use, but much less rapidly than flash memory, which has a capacity of approximately 5,000 write cycles, while PRAM can provide around 100 million cycles. PRAM can offer high performance, especially in applications requiring many write cycles, as it provides fast switching times: a single bit changes without needing to erase blocks first, and it is currently capable of scaling to 40 nm.
Micron Technologies and Samsung are currently racing to bring the first 1 Gbit PCM chip to market. Samsung currently uses a 512 Mbit PRAM-compatible NOR-Flash memory in mobile phones.
FRAM or FeRAM:
The construction of FRAM memory is similar to that of DRAM memory. DRAM consists of a dielectric access transistor and a capacitor-based storage node, while FRAM memory uses a ferroelectric material such as PZT to form a ferroelectric capacitor, which is integrated into the gate body of the transistor.
The crystalline structure of ferroelectric material allows the formation of semi-permanent electric dipoles, which align with the direction of an applied external electric field and retain that polarization when the field is removed. This allows random access to each individual bit for both read and write operations, with binary "0s" and "1s" stored in the two possible polarizations for each cell.
FRAM consumes less power, has a faster write cycle, and can perform many more write-erase cycles than flash memory. Disadvantages include lower storage density, limited storage capacity, and higher costs.
Fujitsu has announced that they are designing FRAM-based devices to replace flash and SRAM memory in industrial, manufacturing, and low-power applications, while Texas Instruments is slowly beginning to offer FRAM-based microcontrollers.
MRAM
memory uses magnetic storage elements formed from two ferromagnetic plates, each capable of maintaining a magnetic field,
separated by a thin insulating layer. The simplest cell structure is the spin valve configuration. One of the plates is a permanent magnet of fixed polarity, and the other has a field that can be changed to match an external field. Bits are stored by representing “1s” with two plates of the same polarity and “0s” with plates of opposite polarity. A grid of these cells forms a memory device. The technology promises the potential replacement not only of Flash, but also of DRAM and SRAM. MRAM, however, is susceptible to interference from the long-term application of torque caused by external DC magnetic fields. Everspin, a spin-off of Freescale Semiconductor, expects to sell MRAM in volumes of millions in 2012, although it is already examining a second-generation MRAM technology called SST-MRAM. This technology replaces the insulating layer with a tunnel barrier layer and uses polarized electrons. The main advantage is the reduction in current required for writing, making the process comparable to the reading process and enabling higher density. However, spin coherence must be maintained, and high-speed operation still requires a higher current. Currently, the technology is expected to be used in devices below 65 nm, although new composite structures are being investigated.
RRAM, or ReRAM
, is based on the electronic switching (induced by current or voltage) of a resistive material between two stable resistive states (high/low). This is achieved through sudden conduction across oxide insulators. RRAM memory switches between the two resistance states through two operations: RESET restores a high-resistance state from a low-resistance state, and SET produces the opposite transition. RRAM memory is scalable below 30 nm, and a study of oxide-based RRAM has even suggested that oxygen movement can occur at the 2 nm level. The IMEC research institute predicts that stacked RRAM devices could appear on the market at 11 nm, with the "SONOS" flash memory serving as an intermediate step at the 17 nm and 14 nm nodes. RRAM memory offers sub-nanosecond switching with minimal power consumption and provides data stability, withstanding high temperatures and the wear generated by cycles. This robustness offers new opportunities in the automotive markets and in embedded applications.
Elpida, for example, has already developed RRAM prototypes with the goal of entering mass production in 2013, with devices offering gigabit capabilities and based on a 30nm process. And there's still no end in sight...
However, none of these four technologies has achieved mass production, as they are currently limited to niche markets. Flash memory is also likely to scale further, at least for a few more generations, so it could be some time before we can say, “The king is dead, long live the king.”.
Author:
By Mark Cundle, RS Components
