Tech Talk iconTech Talk

Flashristors: Getting the Best of Memristors and Flash Memory

Memristors, devices that contain conductors that 'remember' the current that has passed through them by changing their resistance, are being extensively investigated for potential application in memory cells.   

The advantage of memristor memory is that it is nonvolatile, so it does not require continuous power to retain data, as DRAMs do.  Also, memristors have proven to be a possible candidate for multistate memory cells, opening the way for their use in bionic brains. One drawback is that memristors’ repeatability, or the number of times a memristor state can be switched on and off, is limited. 

A team of researchers at Bingöl University in Bingöl and Bilkent University in Ankara, both in Turkey, reported on 12 June in Applied Physics Letters that they’ve developed a 'flashristor,' a device that combines both the properties of a memristor and a flash-memory cell. Unlike existing devices, the memristor effect in their flashristor is not a local effect (such as a change on an atomic scale in crystalline structure), but a distributed effect in a larger area of the device, explains Aykutlu Dâna, one of the paper’s authors.  “This way you can switch on and off with a high precision and repeatability,” he adds.

Dana's team, which has been researching flash memories for some time, realized that a field-effect transistor could behave like a memristor if instead of varying the voltage of the gate via an external terminal, you charge it with electrons that remain trapped by changing the drain voltage only.  The apparent resistance of the channel (which connects the source with the drain) would then depend on the charge in this "floating gate."

This floating gate—called "storage layer" in the paper—needed to have three properties: 1) it had to be able to store electrons; 2) it had to be well insulated so that it retains the charge; and 3) there had to be a path that allows electrons to easily reach this layer.  To create the flashristor device, they used a structure similar to that of a thin-film flash memory. 

The channel consists of a zinc oxide layer connected to the source and drain via aluminum contacts. A 15-nanometer-thick hafnium oxide tunnel barrier allows the passage of electrons from the channel to the storage layer—a 5-nm-thick layer of silicon nitride. The storage layer is separated from the p-silicon substrate (which acts as a gate) by a control layer of aluminum oxide. 

The device differs from a flash memory cell in that the source is connected to the p-silicon gate. When, for example, a negative 5-volt pulse is sent to the drain, the electric field causes electrons from the channel to pass through the tunneling layer and become trapped in the storage layer.  When the pulse stops, the channel "sees" the charge in the storage layer, and its apparent resistance is changed into a different memristor state.  The resistance of the channel can be read out with a 0.1-volt pulse; but this voltage is too low to affect the electron population in the storage layer.  The flashristor can be reset by a positive 5-volt pulse, which gives the electrons in the storage layer sufficient energy to return to the channel through the tunnel barrier.

The researchers demonstrated four memristor states. But the uniformity and repeatability of the charge is better than a few percent, so it’s possible in principle to generate with 10 different states, explains Dâna.

Currently the write and erasing times are too long (about one second), admits Dana, who reiterates that the experiment was just a proof of principle.

Harika Manem, an electrical engineer at SUNY Polytechnic Institute’s Colleges of Nanoscale Science and Engineering, agrees that the device, which is CMOS compatible, needs to record memory states more quickly. “I liked that their analysis and modeling backed up their experimental results, but I don't think it is very applicable for neural circuits or other applications; you need much faster writing times,” she says.   

Improvement should be possible, says Dâna. “We have some preliminary results that show an extremely high performance of graphene as a storage layer,” he says. “The conduction band of graphene sits right in the middle of the valence and conduction bands of the channel, and it can more efficiently capture the tunneling electrons from the channel, so it can be written in very short times. We have some initial devices that can be charged below a microsecond; this is better than conventional flash," says Dâna. But the road to practical applications might still be long. "Maybe if we come up with devices that can be operated with nanosecond write/read pulses, then industry will become interested,” concludes Dâna.

A Second Life for Charge Pumping

An international research group has given new life to charge pumping, a former mainstay of semiconductor testing that had been nearly abandoned when its sensitivity failed to keep pace with ever-shrinking chips.

A handful of single-atom defects in a semiconductor chip—a few holes in the lattice, dangling bonds, an single foreign atom impurity, or even an extra interstitial silicon atom shoe-horned in to disrupt the structure—can degrade the chip’s performance. When they occur at the junction of a transistor gate and the dielectric layer, these “interface traps” badly distort the device’s response as applied electrons fill and drain from these backwater pools.

Finding such defects in metal-oxide field effect transistors quickly and inexpensively is essential to maintaining quality.

In traditional charge pumping, testers apply a square-wave input signal to the transistor gate, cycling the voltage to change the gap bias and generate a signal that can be measured on the far side of the substrate (which is also connected to the source and drain).

In a normal transistor, the output signal echoes the input in a consistent way. In a defective chip, though, the electrons eddying in the interface traps degrade the response. A single input signal won’t find all of the defects, though: Different traps fill and empty at different rates, so the engineers repeat the test with a number of different input signal frequencies.

The test is further complicated by current leaking through the dielectric layer, which adds a leakage current to the charge pumping signal. This was a negligible problem at first; leakage through the relatively thick dielectric layers of earlier transistors was relatively small. But chips got smaller, and so did the charge pumping signal; but the leakage current grew. And about a decade ago, the usable signal vanished almost completely amid the leakage noise. Charge pumping ceased to be a reliable technique for evaluating the most advanced devices.

In IEEE Transactions on Electron Devices, researchers at NIST in Gaithersburg, Md.—with colleagues from Peking University in Beijing, IBM Research in Albany, N.Y., and TSMC in Hsinchu, Taiwan—report on a revision of the technique that both revives charge pumping’s utility and simplifies the measurements.

Traditional charge-pumping applied a square wave input at a single frequency and measured a direct current output. Individual measurements at multiple frequencies were needed, and output data had to be manipulated off-line to subtract out the leakage signal.

NIST researchers Jason T. Ryan and Kin P. Cheung (the corresponding author) and their collaborators have developed an elegant solution that both cancels the leakage current error and generates results directly, without manipulation. They have replaced the series of single-frequency square waves with a frequency modulated input signal that combines two alternating test frequencies in a single input. If the higher first input signal is f1 and the second is f2, the output frequency, f3, is the difference between them (f3=f1-f2, the familiar “beat” of musical notes that are not quite in tune). The combination cancels the leakage current, so that the combined AC output current reflects the charge pumping alone.

In the process, the researchers also found a systematic error in the traditional test. The classical method assumed that leakage current did not vary with the input frequency, while the charge pumping output did. The earlier chip-testers plotted total output signal against the multiple input frequencies, extrapolated the curve to zero frequency, took that value to be the leakage current, and subtracted it from their other results to find the true charge pumping output.

Ryan, Huang, and their collaborators found that while it is technically true that leakage current is independent of the input frequency of an ideal square wave, in the real world, the leakage current does, in fact, change with frequency. Because square waves are, of course, not absolutely square, there is a finite time during which the voltage is rising, and another during which it is falling. These intervals do slightly alter the amount of current flowing through the dielectric.

Illustrations: NIST

This article was edited 21 June 2015 to include NIST’s Kin P. Cheung as corresponding author, rather than the last author, Peking University’s Ru Huang.

Using Light to Activate a Mouse's Happy Memory Protects It from Stress

This is one of the joys of writing about neuroscience: finding a press release image like the one above with the caption, “Cross section of a happy memory.” 

Need a little more explanation? The image shows neurons in a mouse brain that were active when the rodent formed a happy memory. And that memory had therapeutic value: When researchers artificially reactivated those neurons while the mouse was under stress, the mouse exhibited fewer depression-like behaviors. The research was published this week in Nature

There’s still a lot to unpack here: What’s a happy memory for a mouse? Why was it so stressed out? And how can you tell if a mouse is depressed? 

Read More

First CubeSats Planned for Mars

The first interplanetary CubeSats—satellites based on cubes just 10 centimeters wide—will be deployed during the next mission that NASA sends to Mars, the agency says.

CubeSats, whose dimensions are based on the size of a Beanie Baby box one of their inventors found in a shop, are one of the cheapest, most efficient ways to get communications networks into space. Weighing in at just 1 to 10 kilograms, these "nanosatellites" usually pack little more than solar panels, communications equipment, and a few scientific instruments. But now researchers are developing tiny propulsion systems for CubeSats to help them orient themselves, maneuver, and even rocket to new orbits.

Read More

More Light From Metamaterials

A material that manipulates light in unusual ways could lead to a whole variety of exotic devices, including microscopes capable of seeing inside cells, optical circuits for quantum computers, and invisibility cloaks.

Read More

Injectable Electronics Give Neurology a Shot to the Brain

Tiny electronic meshes that can be injected directly into the brain might one day provide control of prosthetic limbs, repair brain damage, or make artificial eyes, according to scientists who have already inserted the devices into the brains of mice.

Read More

Implant Fights Stroke, Tinnitus by Retraining the Brain

human os icon

Houston-based Microtransponder is using an implanted vagus-nerve simulator to turbocharge one of the brain’s most fundamental functions: learning. “What we’re doing with our VNS pairing therapy is trying to reorganize the brain,” says the company’s R&D director, Navzer Engineer.

Vagus nerve stimulation (VNS) involves a device, usually implanted in the body, that sends electric signals up into the brain from a nerve in the neck. Many of the organizations pursuing vagus-nerve-based treatments are targeting the brain’s centers that release neurotransmitters to treat conditions as various as epilepsy, migraine headache, and heart failure. The vagus links to a deep-brain structures, such as the nucleus basalis and the locus coeruleus, which are stuffed full of neurotransmitters like acetylcholine and norepinephrine—key neurochemicals in the cellular mechanisms of learning and memory. The neurotransmitters tell the brain what to learn and when to learn it, says Engineer. Microtransponder’s system is designed to help the brain learn its way around damage.

Read More

5 Materials Innovations for New Medical Devices

human os icon

The next frontier for electronics could lie inside the human body, with sensors that keep track of biomarkers and brain activity, systems to deliver drugs or monitor exercise levels, and communications networks that allow such devices to call on the processing power of your smartphone and send your data to the doctor’s office.

“We’re moving toward a world where rather than going to the doctor once a month and having a measurement made, we’re going to have continuous monitoring,” says Ifor Samuel, head of the Organic Semiconductor Optoelectronics group at the University of St. Andrews, Scotland. Goaran Gustafsson, an expert in printed electronics and bioelectronics at the Swedish research institute Acreo Swedish ICT, says the idea is sort of OnStar for the body, similar to the General Motors system that lets a technician unlock your car remotely or notify the tow truck when you’ve had a breakdown.

Wiring the body could provide more realistic measurements of everything from blood oxygenation and to stress hormone levels, keeping them in the context of everyday life and perhaps providing early warnings of problems. Samuel and Gustafsson were among researchers at the Materials Research Society’s December meeting in Boston who talked of efforts to develop a network of devices that would keep track of people’s health and intervene where necessary. While most of these proposals will take several years to develop, and will require regulatory approval, here are some ideas these researchers are pursuing.

Read More

Powered Prosthetic Legs Work Better by Tracking EMG

Powered prosthetic legs work better when guided by electrical signals generated by the muscles, says a report published today in the Journal of the American Medical Association (JAMA). The findings suggest that bionic legs that rely on mechanical sensors to control movements would be greatly improved by the inclusion of electromyographic (EMG) data and the algorithms that interpret them.

In the study, teams from the Rehabilitation Institute of Chicago and Northwestern University tried out their system on seven people with above-knee amputations. Each participant was outfitted with 9 EMG sensors on their thighs and hips that were connected to a computer. The participants wore a prototype knee-ankle prosthesis powered by 13 mechanical sensors that measured inertia, load, position, angle, acceleration, velocity and torque of the knee and ankle joints. The prosthesis was developed by Michael Goldfarb, a mechanical engineering professor at Vanderbilt University.

The participants were asked to traverse varying terrain—up ramps, stairs, and across level ground. The prostheses relying on the mechanical sensors alone made errors 14.1 percent of the time. But when the EMG data was incorporated, the prostheses made errors only 7.9 percent of the time—or about half as often.

“That’s a lot,” says Levi Hargrove, a research scientist at the Center for Bionic Medicine at the Rehabilitation Institute of Chicago, who led the study. The seven participants were not told when the EMG data was being used, but “every subject figured it out,” said Hargrove. “At the end of the experiment, we would ask: ‘Which of the two conditions did you prefer?’ And they chose the one that used the EMG signals every time.” 

The advance is good news for people with above-knee amputations. Designing prostheses for these kinds of amputations is complex because they require precise coordination of knee and ankle movements. Particularly tricky is transitioning from one type of walking to another—like from walking on flat ground to climbing stairs.

There are only two powered leg prostheses on the market. One, which provides movement in just the knee joint, is made by an Icelandic company called Ossur. The other, called Biom, has a moving ankle; it was developed by Hugh Herr’s group at MIT. Prosthetics that combine powered knee and ankle movement are all still in the prototype stage.

A problem with these mechanical prototypes is that their designs don’t offer the flexibility necessary to accommodate different gaits. The user has to stop moving and make some kind of exaggerated body motion, or use a remote control to tell the leg what to do next—a problem that is frequently awkward and potentially dangerous. 

Hargrove’s team aims to make walking smoother and safer for people with prosthetic legs. “We want them to be able to approach and walk up stairs the same way you and I would,” says Hargrove. 

Arm prostheses are much more advanced than those for the legs. The “Luke Arm” developed by the DEKA Research and Development Corp, for example, harnesses EMG data from the amputee’s remaining arm muscles and interprets them to allow the prosthetic limb carry out multiple, simultaneous movements in the wrist and fingers that allow pinching or gripping. The device received FDA approval last year. Hargrove and his colleagues helped develop the system, and have since commercialized algorithms that can control the Luke Arm or any other arm prothesis that relies on EMG pattern recognition. 

But systems for the leg—particularly for combined knee-ankle devices—have remained elusive. Why the difference? Arms have been a clinical focus for longer than legs. Plus, technological challenges—developing motors and actuators that are strong enough, light enough, and efficient enough to carry people throughout the day without having to recharge batteries—have been difficult to overcome. But that is changing. “All of these innovations are coming together to make these categories of devices available,” says Hargrove. “So now we need to learn to control them as best we can.”

Hargrove and his team, in a collaboration with Vanderbilt and the U.S. Army, are now testing the systems on 15 participants in home settings. “That’s the real test,” says Hargrove. “We’re trying to understand if this is useful for people in the real world.”

Fraunhofer Optics Could Make Augmented Reality Specs Thinner

The list of companies with an interest in augmented reality eyewear has grown long over the years and includes some of the biggest names: Microsoft, Google, and Sony. However, consumer acceptance has been lackluster, in part because these specs are still a bit bulky.

Researchers led by Peter Schreiber at the Fraunhofer Institute for Applied Optics and Precision Engineering IOF in Jena, Germany, say they’ve created demonstration models of  data glasses that would be much smaller and less obtrusive than the specs available now.

The key to their design is the use of a glass waveguide for light, both for the transmission and display of images.  "We have a very slim design for the eye-piece optics, which fits very well with the geometry of a wave guide," says Schreiber. A simple, elongated rectangular sheet of glass serves both as a wave guide and as the support for a small, transparent display of 8 by 15 millimeters near the eye.

A two-megapixel microdisplay produces the image at one end of the glass sheet. This image is coupled into the glass sheet by a diffraction grating—a thin microlattice deposited on the glass waveguide as a thin, transparent plastic film on the glass.  This image travels through the glass sheet and is coupled out of the wave guide at the other end close to the eye of the user by a similar microlattice.  Held close to the eye it can be viewed via an array of one-mm microlenses.

Unfortunately, using a waveguide for transmitting the image limits its definition on the display to about 800 x 600 pixels. "We have a very slim design, and we pay for this slimness in terms of projected pixels," says Schreiber.  But this would not drastically limit the application of the specs, he claims.  "If you want to use a simple navigation system, which only says, '500 meters to the right,' or …. 'there is a call on your smartphone,' or for simple graphics in a repair manual, then it will work well." 

Up to now the researchers have produced two demonstration prototypes of their data glasses. One is a real near-to-eye display with a micro imager, but it isn’t see-through. The other system cannot display moving images, but is see-through, says Schreiber. "What we have to do in the future is to put both systems together, with a real micro imager and a waveguide for the see-through option,"he says.

An additional benefit of the Fraunhofer system is that by manipulating the images—image processing—one can correct for a user’s farsightedness. Your smartphone is connected wirelessly to your display, and you will have a smartphone app that will allow you to enter your degree of vision defect, such as +1 or +2, says Schreiber. The smartphone will then process the images, allowing you to see a sharp display, says Schreiber.  He argues that moving most of the image processing power and operating software to a smartphone allows a more compact structure for the data glasses, and a longer battery life.  

The technology they are developing might also be of use in several optical instruments, such as microscopes and micromanipulators, where transparent displays could be used for placing cross hairs or a controllable cursor in the view. 

 "For future optical solutions, we are now looking for partners, like startups, to develop products with them," says Schreiber.


Tech Talk

IEEE Spectrum’s general technology blog, featuring news, analysis, and opinions about engineering, consumer electronics, and technology and society, from the editorial staff and freelance contributors.

Newsletter Sign Up

Sign up for the Tech Alert newsletter and receive ground-breaking technology and science news from IEEE Spectrum every Thursday.

Load More