Researchers have developed a way around one of the key failings of ferroelectric RAM (FRAM) that has prevented its commercialisation: a method of reading its contents without destroying it.
A stumbling block to replacing DRAM and NAND flash with FRAM components has been surmounted, but challenges still remain.
With traditional computer circuitry rapidly approaching the limits of what it can do - fans of Moore's Law, the Intel co-founder's observation that the number of transistors on a chip doubles roughly every 18 months, will be familiar with the difficulties inherent in shrinking silicon-based circuits much further - researchers are turning their attention to alternative methods of improving computer performance or capacity. Memristors
are one path, DFG-FETs
another, and FRAM a third.
FRAM, also known as FeRAM, is based on a polarised ferroelectric layer, cells of which can be flipped between storing a zero and a one by the application of voltage. FRAM has numerous advantages over traditional dynamic RAM (DRAM), not least of which is that it is non-volatile: while DRAM requires a constant refresh cycle to keep its contents, FRAM is happy to store its data without the application of any power. Unlike other non-volatile memories like NAND flash, it's also extremely fast - to the point where it is being considered as a candidate for 'universal memory,' where the distinction between RAM and mass storage is erased.
Sadly, while the memory might be non-volatile in terms of retaining contents without power it has another aspect that has given attempts to commercialise it in anything other than small niche markets a headache: reading the data back from the memory erases it. As a result, while the constant refresh cycles of DRAM are avoided, an FRAM system needs to have every bit that is read immediately rewritten back again - slowing the system down and, more importantly, impacting longevity and reliability.
Ramamoorthy Ramesh and Junling Wang, researchers from the University of California at Berkeley and Nanyang Technological University in Singapore claim to have solved the issue by creating a cheap and fast non-destructive read system based on visible light. Using a prototype FeRAM circuit based on bismuth ferrite, exploiting a property of the material discovered in 2009 by Rutgers University researchers whereby it creates a voltage as a result of light exposure, the team was able to read differing voltages depending on the polarity of the bit - reading back the stored data without destroying it.
While non-destructive read systems for FRAM have been developed in the past, they typically use expensive components - the first non-destructive FRAM read system, developed at NASA's Jet Propulsion Laboratory in 1991, used pulses from a powerful ultra-violet laser which the team admitted at the time 'would require a reduction in the [output] power by about an order of magnitude
' before it could be used commercially.
The team's prototype shows real promise: according to figures released by the team as part of the paper's publication in the Nature journal
, the bismuth ferrite FERAM circuit is capable of performing read and write operations in around 10 nanoseconds with a write voltage draw of 3V. By contrast, NAND flash components typically take thousands of times longer and require up to 15 volts.
Sadly, there's a catch to commercialisation: the team's prototype strips measure 10 micrometres, significantly larger than the nanometre-scale components used in modern circuitry. While Ramesh claims that there is no fundamental reason why the process could not shrink down to 22nm or smaller nodes, he does admit that there will be challenges along the way - meaning DRAM and NAND flash are far from dead yet.