Research New universal memory is fast, energy-efficient, stable, and durable

Related Vendor

Wuxi InfiMotion Technology Co., Ltd.

Stanford researchers have developed a new phase-change memory that could help computers process large amounts of data faster and more efficiently.

Our computers need to process ever-increasing amounts of data to, for example, accelerate the discovery of drugs, improve weather and climate predictions, train artificial intelligence, and much more. To keep up with this demand, we need faster and more energy-efficient computer storage now more than ever.

Stanford researchers have shown that a new material could make phase-change memory - which relies on the switch between high and low resistance states to generate the ones and zeros of computer data - an improved option for future AI and data-centric systems. Their scalable technology, recently described in 'Nature Communications', is fast, power-efficient, stable, durable, and can be manufactured at temperatures compatible with commercial production.

"We're not just improving a single metric like endurance or speed, but several parameters at once," says Professor Eric Pop from Stanford University. "This is the most realistic and industry-friendly thing we have built in this field. I would like to see it as a step towards a universal memory."

A faster NV memory

Today's computers store and process data in different locations. Volatile memory - which is fast but fleeting when the computer is turned off - is responsible for processing, while non-volatile memory - which is not as fast, but can store information without a constant power supply - is responsible for long-term data storage. Moving information between these two storage locations can lead to bottlenecks as the processor waits for large amounts of data to be retrieved.

"Moving data back and forth is very energy consuming, especially with today's workloads," said Xiangjin Wu, co-author of the study and a Ph.D. student, supervised by Professor Pop and Professor Philip Wong at the School of Engineering. "With this kind of memory, we really hope to bring memory and processing closer together, ultimately in one system, so it uses less energy and time."

There are many technical hurdles on the path to an effective, commercially viable universal memory, which allows both long-term storage and fast, energy-efficient processing without compromising other system parameters.

But the new phase-change memory developed in Pop's lab has come the closest yet. The researchers hope it will inspire further development and adoption as universal memory.

The memory is based on GST467, an alloy of four parts germanium, six parts antimony, and seven parts tellurium, which was developed by colleagues at the University of Maryland. Pop and his colleagues have found ways to arrange the alloy between several other nanometer-thin materials in a superlattice, a layer structure with which they have previously achieved good results in non-volatile memories.

"The unique composition of GST467 gives the memory a particularly high switching speed," said Asir Intisar Khan, who completed his Ph.D. in Pop's lab and is a co-author of the work. "Integration into the superlattice structure in nano-devices allows for low switching energy, good endurance, very good stability, and makes it non-volatile - it can maintain its state for 10 years or longer."

Setting a new benchmark

The GST467 superlattice sets several important benchmarks. Phase change memory can sometimes drift over time - essentially, the value of the ones and zeros can slowly shift - but their tests showed that this memory is extremely stable. Moreover, the memory works with less than 1 V supply voltage, which is the goal of a power-saving technology, and is significantly faster than a typical solid-state drive.

"Some other NV memories might be a bit faster, but they operate at a higher voltage or with a higher energy consumption," according to Pop. "With all these computing technologies, there are compromises between speed and energy. The fact that we can switch in a few tens of nanoseconds while operating with less than one volt is a significant achievement."

The superlattice also accommodates a large number of memory cells in a small space. The researchers have reduced the memory cells to a diameter of 40 nm - less than half the size of a coronavirus. This isn't quite as complex as it could be, but the researchers are looking for ways to compensate for this by vertically stacking the memory, which is possible thanks to the low manufacturing temperature of the superlattice and the production techniques used.

"The manufacturing temperature is far below what is needed," says Pop. "There is talk of stacking memory in thousands of layers to increase density. This type of memory could enable such 3D layers in the future." (mbf)

* Henning Wriedt is a freelance specialist author.