A survey of nanoengineered memory solutions
01/01/2005
It is tempting to dismiss much of the cutting-edge research being conducted in the nanomemory sector as having little commercial application. However, there are fundamental weaknesses with the technologies currently in use, and until recently, there had been no obvious way to overcome them. Examples are the low access times of bulk storage, the continual refreshing required for dynamic random-access memory (DRAM), and the fact that both DRAM and static RAM (SRAM) are volatile. Flash falls well short of requirements for mobile computing/communications applications with regard to read/write speed and lifetime (number of read/write operations).
These existing limitations represent opportunities for alternative technologies and persist even if scaling continues without problems, which is questionable. Floating-gate flash designs are proving especially difficult to scale, SRAM is looking increasingly vulnerable to soft errors, and DRAM capacitor issues will continue to arise with scaling despite a respite from the use of trench capacitors. Some of these problems are potentially surmountable using the new technologies summarized here.
MRAM
One of the most widely discussed nanoengineered memories is magnetic RAM (MRAM), a technology being pursued by Freescale, Cypress, Honeywell, IBM and Infineon (jointly), NEC and Toshiba (jointly), TDK, Samsung, Sharp, Philips, and more.
The most promising commercial MRAM developments use the effect of spin alignment on a tunneling current perpendicular to two films that can be spin-aligned or unaligned. Having the current flow perpendicular to the films rather than parallel is architecturally simple, compatible with standard lithography, and easily scalable to a dense memory structure.
MRAM’s advantages are that it is as fast as DRAM or SRAM and can achieve similar capacities, yet it is nonvolatile. When compared to flash, read/write times are 1000-10,000× faster (the latter for write times). To date, the biggest obstacles have been scaling while retaining spin orientation, and cost, but several companies appear to be gearing up for commercialization.
Nonoptical phase-change media
A class of ceramic alloys called chalcogenides has been investigated for years by many of the large semiconductor manufacturers, including Samsung, Intel, and STMicroelectronics. For a time, this technology seemed to be in limbo but has recently gained a new lease on life.
With this technology, the phase of the material, and thus data, is measured through changed conductivity. The phase change is achieved using an electrical pulse from a transistor. The approach promises to be inexpensive and, in theory, relatively fast (read/write times both <100 nsec), and with excellent nonvolatility, scalability, and power consumption.
Molecular memory
There are several approaches to using molecules as memory. The molecules can be induced into conformational (shape) changes that alter their resistance, which can then be measured, or altered optical properties can be measured. Molecules can also be used as devices to hold charge, whereby there is no conformational change. This is the most promising for commercialization and is being developed by ZettaCore using porphyrin molecules. The biggest advantage here is that the charge storage concept is analogous to that of DRAM and is easily integrated with CMOS technology.
Rotaxanes have ring elements on a central chain that will change configuration under an applied field. Hewlett-Packard has demonstrated a 64bit, very dense memory based on rotaxanes at the intersections of a nanowire crossbar structure. Another molecular memory candidate is chiropticene, favored by CALMEC and offering the advantage of being optically readable. Spanish researchers have patented an optically activated molecular switch based on dendrimers. Structures based on biomolecules such as proteins and DNA offer some of the greatest versatility in this area but are not known for their resilience. Matsushita is working on a memory technology based on protein aggregation that the company claims would equal the best silicon memories in density but consume 1/100 of the power.
Nanotube RAM
A particularly promising approach to nonvolatile memory is being pursued by only one company - Nantero. The technique is based on carbon nanotubes and dubbed NRAM. Imagine one nanotube at right angles to another and with a gap between them; currents can cause electrical fields to make them attract and bend until they make contact. Once in contact, they are held there by Van der Waals forces until an applied current separates them again. When the nanotubes are in contact, they are out of contact with adjacent electrodes and when they separate, they again make contact with the electrodes. Reading a cell consists of detecting the current between the nanotubes and the electrode. Since carbon nanotubes are highly resilient structures, NRAM is potentially highly stable and nonvolatile.
MEMS-based systems
By definition, MEMS-based memory systems are not nanotechnology, but several promising memory/storage solutions operate at the nanoscale and are simply MEMS-driven.
The most famous of these is the IBM Millipede. IBM’s approach uses an array of atomic force microscope probes to read and write (using heat) indentations on a polymer substrate. The system is low-power, nonvolatile, and shock- and radiation-resistant, and can achieve high read/write speeds through massive parallelism (arrays of >4000 probe tips have been demonstrated).
Other scanning probe techniques are applied to memory technologies used at other levels, such as magnetic or phase-change approaches. HP has worked on using the probes to write to phase-change materials, and groups at Carnegie Mellon U. and the U. of Twente in the Netherlands have developed memory based on using probes with magnetic materials. Nanochip is using a phase-change version of the technology, and Cavendish Kinetics has a version that depends on a tiny cantilever (a flexible beam attached at one end) being bent by an electric charge until it meets another surface, to which it sticks through surface effects. The Cavendish technology is extremely robust and promises to scale well, with access speeds already around DRAM level.
Polymers
There is not always a clear line between polymer-based memory and molecular memories. The polymer memory that Coatue was developing before it was acquired by AMD, for example, used the crossbar architecture popular with molecular memories. This technology promises high switching and high densities, thanks to storage of multiple bits/cell and 3D stacking.
HP also has developed a polymer memory technology that has one substantial virtue - it’s very cheap. It is also low-power and potentially high-density, using 3D packing. It is only a write-once technology, however, so it would not be a flash replacement, but might, for example, hold movies.
Other approaches
Several other advanced memory solutions are currently in various phases of commercialization. Practical nano floating-gate memory has been proposed that can be married with traditional MOSFETs and can also be implemented with carbon nanotubes or nanowires. Nitinol, the shape memory alloy, has been shown capable of creating an extremely dense phase-change memory using electron beams to write rather than heat. Yet another nanomemory approach is ONO-type flash memory that uses a silicon nitride layer between two oxide layers (hence, ONO) to store charge.
Conventional semiconductor memories have come a long way, from densities of <1kbit in 1972 to 1Gbit in 2002. Such memories will play an important role in the electronics industry for decades. Over the next few years, however, as the demand ramps up for memories that are out of the capabilities of garden-variety CMOS, a new market for nanomemories seems sure to emerge.
Reference
This article is based on a report published by NanoMarkets LC (visit www.nanomarkets.net for details).
For more information, contact Lawrence Gasman, principal analyst, at NanoMarkets LC, P.O. Box 7267, Charlottesville, VA 22906; ph 434/984-0245, ext. 11, e-mail [email protected].