IBM Research chart compares memory technologies. Credit: IBM Research
Two new studies serve as bookend demonstrations of the near-term potential for phase-change memory materials for use as nonvolatile data storage devices, and the long-term potential for these materials to change fundamentally what a computer memory is. A third study just out considers the efficiency of the crystalline to amorphous transformation on the lattice scale.
In an important step toward near-term adoption of PCMM technology for nonvolatile data storage, IBM Research, Zurich, Switzerland has demonstrated a device capable of stable multibit storage over an extended time.
Longer term, a research group at the University of Exeter in the United Kingdom has shown that PCMMs can store and process data simultaneously. That is, PCMM has the potential to mimic the human brain’s neurons and synapses and have ability to learn and process information.
A press release from IBM describes work that breaks down two barriers to using PCMM: the ability to store multiple data bits and the ability of the material to reliably store data for extended time periods.
The advantages of PCMM over flash memory are volume, speed and durability. The IBM press release says that PCMMs can write and retrieve data 100 times faster than flash and can endure 10 million or more write cycles.
By comparison, today’s consumer-grade flash memories last about 3,000 write cycles (which tend to be more than enough) and enterprise-grade flash memories are good for upwards of 30,000 write cycles. The real promise of PCMM materials is for the enterprise-grade (nonconsumer applications like business or research IT). Durable for over 300 times more write cycles, combined with the ability to hold data when the power is off, PCMM would allow servers and computers to boot instantly and to handle very large volumes of data quickly. An earlier post described a prototype device that was demonstrated at the Device Automation Conference in June and summarized the computing advantages of PCMM memories.
Applying a voltage, current or optical pulse causes the phase change to occur, and the extent of the phase change is proportional to the applied stimulus, and this, in turn, is a way to control the resistance of the cell. Taking advantage of the proportionality of the resistance, the IBM team was able to store multiple bits—four in this case—in a single cell by manipulating the applied voltage.
Getting the data in is only valuable if the data can be gotten out later. PCMMs have a tendency to “resistance drift” caused by structural relaxation of the amorphous material, which increases resistance and, therefore, errors during read-out of the data. The press release explains that “…on average, the relative order of programmed cells with different resistance levels does not change due to drift,” which allowed the IBM team to solve the problem with an “advanced modulation coding technique that is inherently drift tolerant.”
An IBM test device is storing bits in a subarray of 200,000 cells in a PCMM test chip. The retention test has been running for over five months, thus demonstrating a volume and stability that are of practical use.
On a smaller scale but with potentially bigger implications, David Wright’s group at the University of Exeter in the UK used a single cell of PCMM to demonstrate that the material could store and process information simultaneously, similar to the way neurons and synapses do in the human brain. They also showed the ability perform the simple arithmetic operations of addition, subtraction, multiplication and division. Results are published in Advanced Materials, “Arithmetic and Biologically-Inspired Computing Using Phase-Change Materials” (doi: 10.1002/adma.201101060)
As explained in a press release from the Exeter, the processing and memory functions for current technology are separate, “resulting in a speed and power ‘bottleneck’ caused by the need to continually move data around.” Multicore processors are an effort to get around the power and speed lost to the bottleneck. The human brain, however, composed of neurons and synapses, does not distinguish between processing (i.e., computation) and memory.
In an interview published by The Engineer, Wright, a professor at Exeter, said “We’ve shown that phase-change materials have a natural accumulation and threshold property, which makes them a good candidate for simple implementation of a hardware neuron.” Thus, PCMM components may be able to be connected in “networks via structures akin to synapses, potentially opening up an entirely novel way of computing.”
The Exeter group studied GeSbTe and AgInSbTe compounds. The IBM press release did not specify the material, but a chalcogenide composition is assumed.
Finally, a just-published paper in Nature Nanotechnology also addresses PCMM technology by considering the nature of the amorphous state (see Simpson, et al, “Interfacial phase-change memory,” doi: 10.1038/nnano.2011.96). In an earlier post we reported on work that confines the PCMM in a mesoporous SiO2 scaffold to reduce the power needed for the phase change. Similarly, this paper aims to reduce the energy needed to “switch” the material by limiting the movement of atoms to a single dimension, which greatly reduces losses to entropy. By aligning the c-axis of a hexagonal Sb2Te3 layer in the <111> direction of a cubic GeTe layer in a superlattice structure, germanium atoms can switch sites at the interface of the layers. The abstract says they have demonstrated “interfacial phase-change memory data storage devices with reduced switching energies, improved write–erase cycle lifetimes and faster switching speeds.”
The IBM press release says PCMM nonvolatile memories could start replacing flash in the next five years. The basic research coming out of labs like Wright’s, Simpson’s and Chen’s will be the foundation of second and third generation PCMM devices.