Memory is crucial to all computing devices, both for long-term storage of data and for short-term storage whilst information is being processed.
Currently, different technologies are used for different types of memory, as the properties of each type of memory are quite restrictive.
Current Memory Technologies
SRAM (Static Random Access Memory) is mainly used in high performance embedded computing, and in cache memory for processors and hard drives, where its high speed and low energy consumption is helpful. It is very costly, however, and has a very low density compared to other forms of memory.
DRAM (Dynamic Random Access Memory) is also quite fast, and much more dense and cheaper than SRAM, making it the current choice for the main memory banks in computers, shuttling information between the storage drives and the processor.
Flash memory is used where permanent storage is required - DRAM requires power to maintain the arrangement of 1s and 0s on the chip, but a Flash drive is non-volatile, and so will store data indefinitely with or without power. It is relatively cheap and high-density, but is not fast enough for RAM applications. The properties that keep the data stored on a flash chip stable for up to 10 years also means that a large amount of energy is required to write to the chip, slowing down the process. Writing data also damages the flash chip, limiting its useful lifetime.
.jpg)
Figure 1. RAM, or Random Access Memory, comes in lots of different forms. DRAM, pictured, is too slow for high-performance applications, and cannot store data without a constant power source. It is used in computers to shuttle data between the hard drive and the processor cache.
Universal Memory
Semiconductor manufacturers are now competing to produce "universal memory" technologies, which combine the benefits of each of these technologies. The primary aim is a memory with the access speed of SRAM, but with the non-volatility of Flash. There are several potential candidates, explored in more detail below, which are likely to become commercially competitive with current technologies within the next five to ten years.
Another driver for developing these new technologies is to keep up with the exponential progression of Moore's Law. The feature size in silicon-based integrated circuits has halved roughly every two years since the 1960s, but physical limits to this progression are within sight. Many of the universal memory technologies which are being explored have the capability to be scaled down beyond the limits of silicon CMOS circuits.
Carbon Nanotube RAM (CNT RAM)
Carbon nanotubes (CNTs) have great potential as the basis for memory chips. their small size and unique dimensionality allow for interactions between their electrical and mechanical properties which can be used to design fast, dense, and non-volatile data storage devices.
Whilst many CNT-based memory designs have been proposed, the difficulty in producing the nanotubes with a sufficient purity and quality, and with integrating the nanomaterials with current semiconductor fabrication techniques, has prevented their widespread adoption.
.jpg)
Figure 2. Carbon Nanotubes have unique electronic properties that could be used to make highly efficient, fast, non-violative memory chips. However, there are many challenges in bringing the technology to market - mainly manufacturing the nanotubes in high enough purities.
Phase Change RAM (PCRAM)
In 2011, IBM demonstrated a breakthrough in Phase Change RAM (PCRAM), which has been in development as a potential universal memory technology for some time. There will no doubt still be difficulties in translating the technology to a large-scale fabrication process, but the properties are very promising. In June 2012, IBM announced a deal with SK Hynix to take the commercialization of this technology further.
Phase Change memory is based on a special material that has two possible phases - crystalline and amorphous - and can be switched between the two phases using a short electrical pulse. The write speed is around 100 times faster than currently available flash memory, although extra operations are required to check for write errors and correct for drift.
Magnetoresistive RAM (MRAM)
Magnetoresistive technology is a mature technology, which is behind modern high-density hard drives. There has been a recent research drive to adapt this technology to a higher speed, non-volatile solid-state memory. The main challenge to this is to create a large, high-density array of magnetic tunnel junctions, which are used to write to the storage layer. Hard drives contain just one of these, whereas an MRAM chip would need one for each bit of stored information.
Because it based on well-known technology, MRAM is hotly tipped as a candidate for the first commercial universal memory. Companies such as Samsung, Toshiba, IBM, Hitachi, and Motorola are all involved in the development of MRAM.
Quantum Dot RAM (QD RAM)
Quantum Dot RAM uses 3nm-wide spots of semiconductor material, called quantum dots, embedded in a layer of insulating material and covered with a metal film. This structure forms an array of transistors, which are used to store data by changing the state of each quantum dot using a millisecond laser pulse.
This technology is highly promising, as it can achieve read/write speeds up to hundreds of times faster than existing types of memory, and is also reasonably easy to integrate into existing manufacturing processes, as the chips can still be constructed from silicon. There will be challenges in scaling up the process, but not to the same extent as with implementing a totally new material like carbon nanotubes.
.jpg)
Figure 3. Quantum dots are tiny crystals containing just a few hundred atoms. Memory-based on quantum dots could achieve much higher storage densities than existing technologies and would have a vastly longer lifetime.
Conclusions
The main challenges to all of these technologies is to get them to a stage where they can be manufactured affordably, preferably using minimally adapted existing equipment. They must also compete with flash memory on price, speed and data density. In the five to ten years it could take to get these new technologies to market, the flash will also have advanced considerably, so the real requirements for any new replacement technology are very high.
Sources and Further Reading
- "Toward a Universal Memory" - Akerman, Science, 2005. DOI: 10.1126/science.1110549
- NRAM™ - Nantero
- "The Emergence of Practical MRAM" - Barry Hoberman, Crocus Technology
- "Fast programming metal-gate Si quantum dot nonvolatile memory using green nanosecond laser spike annealing" - Lien et al, Appl. Phys. Lett. 2012. DOI: 10.1063/1.3700729
- "106 years extrapolated hole storage time in GaSb/AlAs quantum dots" - Marent et al, Appl. Phys. Lett. 2007. DOI: 10.1063/1.2824884