Memristor Memory Readied for Production
HP has begun testing samples of a new kind of nonvolatile memory based on memristors–circuit elements that are much smaller than the transistors used in flash memory. The company plans to introduce the first commercial memristor memory product in three years’ time.
HP expects its memristor memory technology to scale better than flash and hopes to offer a product with a storage density of about 20 gigabytes per square centimeter in 2013–double the storage that flash is expected to offer at that time. The move will be an important testing ground for memristors; the reliability and performance of these components, first made at HP Labs in 2008, remain largely unproven.
R. Stan Williams, a senior fellow at HP and director of the company’s information and quantum systems lab, says his group is testing the first batch of sample memristor memory devices made at an undisclosed semiconductor fab. The sample memristor arrays are being built on standard 300-millimeter silicon wafers.
Memristors are nanoscale devices with a variable resistance and the ability to remember their resistance when power is off. HP fabricates them using conventional lithography techniques: laying down a series of parallel metal nanowires, coating the wires with a layer of titanium dioxide a few nanometers thick, and then laying down a second array of wires perpendicular to the first. The points where the wires cross are the memristors, and each can be as small as about three nanometers. This cross-bar structure also makes it possible to pack memristors in very dense arrays.
Both flash and memristor memory are nonvolatile, meaning they hold on to data even when power is cut off. Flash has some limitations, though. It can only withstand about 100,000 data-writing cycles, and, like all devices based on silicon transistors, it will come up against physical limits as it’s scaled to make more storage-dense memory devices. Williams says that memristor memory can withstand up to about a million read-write cycles in lab tests. “We will be able to scale faster and farther than flash because the memristor is a very simple structure, and it can be stacked,” Williams says.
Other researchers are cautiously optimistic about memristors’ promise. While silicon’s material properties are well known, those of the materials used to make Williams’s memristors are not–at least so far.
“The fundamentals of why these metal oxides switch the way they do are not well understood,” says Curt Richter, leader of the NanoElectronic Device Metrology project at the National Institute of Standards and Technology (NIST) in Gaithersburg, MD. A better understanding of the fundamental material properties of the metal oxides used to make the memristors will be critical to ensuring that chips with billions of the devices operate reliably over as long as 10 years.
Transferring the technology to fabrication facilities could go a long way toward filling that knowledge gap. “Once you have the fab, it’s a completely new game,” says Dmitri Strukov, professor of electrical and computer engineering at the University of California, Santa Barbara, who is developing memristors in his lab.
It could also help efforts to develop memristor logic circuits, says Richter. Memristors have been the subject of much interest because, in theory, they’re capable of activity that’s analogous to what happens in a synapse in the human brain. So far, however, all the experimental demonstrations of memristors have been accomplished by forcing them to behave more like transistors. Instead of switching between hundreds of states, these memristors have been made to switch between two states with a high and low resistance–a digital zero and one.
This week, in the journal Nature, Williams and colleagues reported a major step forward for memristor logic with the fabrication of circuits capable of full Boolean logic. The circuits are still digital, but Williams says his team has “shown that anything that can be calculated on silicon can be done with memristors,” and in a smaller space. Demonstrating digital logic with the devices is an important first step toward more exotic computing, says Strukov.
The memristor circuits reported in Nature are also capable of both memory and logic, functions that are done in separate devices in today’s computers. “Most of the energy used for computation today is used to move the data around” between the hard drive and the processor, says Williams. A future memristor-based device that provided both functions could save a lot of energy and help computers keep getting faster, even as silicon reaches its physical limits.
For now, though, the company will work to overcome potential manufacturing challenges that arise as it develops memristors for nonvolatile memory. Memristors are passive devices that must be built on top of traditional silicon transistors that serve to introduce power into the system. This complexity could be a hurdle, says Pinaki Mazumder, professor of electrical engineering and computer science at the University of Michigan. “As you introduce more [lithography] masks, it could have a negative effect on yields, because your chance of errors increases,” he says.
In spite of these challenges, Williams says it’s time for memristors to scale up. “Our lab results have been good, and it’s time to test memristors in the fab.”
Geoffrey Hinton tells us why he’s now scared of the tech he helped build
“I have suddenly switched my views on whether these things are going to be more intelligent than us.”
Meet the people who use Notion to plan their whole lives
The workplace tool’s appeal extends far beyond organizing work projects. Many users find it’s just as useful for managing their free time.
Learning to code isn’t enough
Historically, learn-to-code efforts have provided opportunities for the few, but new efforts are aiming to be inclusive.
Deep learning pioneer Geoffrey Hinton has quit Google
Hinton will be speaking at EmTech Digital on Wednesday.
Get the latest updates from
MIT Technology Review
Discover special offers, top stories, upcoming events, and more.