Nanoscience and nanotechnology

Subterms

    More stories

    • 150 Shares179 Views

      in Energy

      New hardware offers faster computation for artificial intelligence, with much less energy

      by

      As scientists push the boundaries of machine learning, the amount of time, energy, and money required to train increasingly complex neural network models is skyrocketing. A new area of artificial intelligence called analog deep learning promises faster computation with a fraction of the energy usage.

      Programmable resistors are the key building blocks in analog deep learning, just like transistors are the core elements for digital processors. By repeating arrays of programmable resistors in complex layers, researchers can create a network of analog artificial “neurons” and “synapses” that execute computations just like a digital neural network. This network can then be trained to achieve complex AI tasks like image recognition and natural language processing.

      A multidisciplinary team of MIT researchers set out to push the speed limits of a type of human-made analog synapse that they had previously developed. They utilized a practical inorganic material in the fabrication process that enables their devices to run 1 million times faster than previous versions, which is also about 1 million times faster than the synapses in the human brain.

      Moreover, this inorganic material also makes the resistor extremely energy-efficient. Unlike materials used in the earlier version of their device, the new material is compatible with silicon fabrication techniques. This change has enabled fabricating devices at the nanometer scale and could pave the way for integration into commercial computing hardware for deep-learning applications.

      “With that key insight, and the very powerful nanofabrication techniques we have at MIT.nano, we have been able to put these pieces together and demonstrate that these devices are intrinsically very fast and operate with reasonable voltages,” says senior author Jesús A. del Alamo, the Donner Professor in MIT’s Department of Electrical Engineering and Computer Science (EECS). “This work has really put these devices at a point where they now look really promising for future applications.”

      “The working mechanism of the device is electrochemical insertion of the smallest ion, the proton, into an insulating oxide to modulate its electronic conductivity. Because we are working with very thin devices, we could accelerate the motion of this ion by using a strong electric field, and push these ionic devices to the nanosecond operation regime,” explains senior author Bilge Yildiz, the Breene M. Kerr Professor in the departments of Nuclear Science and Engineering and Materials Science and Engineering.

      “The action potential in biological cells rises and falls with a timescale of milliseconds, since the voltage difference of about 0.1 volt is constrained by the stability of water,” says senior author Ju Li, the Battelle Energy Alliance Professor of Nuclear Science and Engineering and professor of materials science and engineering, “Here we apply up to 10 volts across a special solid glass film of nanoscale thickness that conducts protons, without permanently damaging it. And the stronger the field, the faster the ionic devices.”

      These programmable resistors vastly increase the speed at which a neural network is trained, while drastically reducing the cost and energy to perform that training. This could help scientists develop deep learning models much more quickly, which could then be applied in uses like self-driving cars, fraud detection, or medical image analysis.

      “Once you have an analog processor, you will no longer be training networks everyone else is working on. You will be training networks with unprecedented complexities that no one else can afford to, and therefore vastly outperform them all. In other words, this is not a faster car, this is a spacecraft,” adds lead author and MIT postdoc Murat Onen.

      Co-authors include Frances M. Ross, the Ellen Swallow Richards Professor in the Department of Materials Science and Engineering; postdocs Nicolas Emond and Baoming Wang; and Difei Zhang, an EECS graduate student. The research is published today in Science.

      Accelerating deep learning

      Analog deep learning is faster and more energy-efficient than its digital counterpart for two main reasons. “First, computation is performed in memory, so enormous loads of data are not transferred back and forth from memory to a processor.” Analog processors also conduct operations in parallel. If the matrix size expands, an analog processor doesn’t need more time to complete new operations because all computation occurs simultaneously.

      The key element of MIT’s new analog processor technology is known as a protonic programmable resistor. These resistors, which are measured in nanometers (one nanometer is one billionth of a meter), are arranged in an array, like a chess board.

      In the human brain, learning happens due to the strengthening and weakening of connections between neurons, called synapses. Deep neural networks have long adopted this strategy, where the network weights are programmed through training algorithms. In the case of this new processor, increasing and decreasing the electrical conductance of protonic resistors enables analog machine learning.

      The conductance is controlled by the movement of protons. To increase the conductance, more protons are pushed into a channel in the resistor, while to decrease conductance protons are taken out. This is accomplished using an electrolyte (similar to that of a battery) that conducts protons but blocks electrons.

      To develop a super-fast and highly energy efficient programmable protonic resistor, the researchers looked to different materials for the electrolyte. While other devices used organic compounds, Onen focused on inorganic phosphosilicate glass (PSG).

      PSG is basically silicon dioxide, which is the powdery desiccant material found in tiny bags that come in the box with new furniture to remove moisture. It is studied as a proton conductor under humidified conditions for fuel cells. It is also the most well-known oxide used in silicon processing. To make PSG, a tiny bit of phosphorus is added to the silicon to give it special characteristics for proton conduction.

      Onen hypothesized that an optimized PSG could have a high proton conductivity at room temperature without the need for water, which would make it an ideal solid electrolyte for this application. He was right.

      Surprising speed

      PSG enables ultrafast proton movement because it contains a multitude of nanometer-sized pores whose surfaces provide paths for proton diffusion. It can also withstand very strong, pulsed electric fields. This is critical, Onen explains, because applying more voltage to the device enables protons to move at blinding speeds.

      “The speed certainly was surprising. Normally, we would not apply such extreme fields across devices, in order to not turn them into ash. But instead, protons ended up shuttling at immense speeds across the device stack, specifically a million times faster compared to what we had before. And this movement doesn’t damage anything, thanks to the small size and low mass of protons. It is almost like teleporting,” he says.

      “The nanosecond timescale means we are close to the ballistic or even quantum tunneling regime for the proton, under such an extreme field,” adds Li.

      Because the protons don’t damage the material, the resistor can run for millions of cycles without breaking down. This new electrolyte enabled a programmable protonic resistor that is a million times faster than their previous device and can operate effectively at room temperature, which is important for incorporating it into computing hardware.

      Thanks to the insulating properties of PSG, almost no electric current passes through the material as protons move. This makes the device extremely energy efficient, Onen adds.

      Now that they have demonstrated the effectiveness of these programmable resistors, the researchers plan to reengineer them for high-volume manufacturing, says del Alamo. Then they can study the properties of resistor arrays and scale them up so they can be embedded into systems.

      At the same time, they plan to study the materials to remove bottlenecks that limit the voltage that is required to efficiently transfer the protons to, through, and from the electrolyte.

      “Another exciting direction that these ionic devices can enable is energy-efficient hardware to emulate the neural circuits and synaptic plasticity rules that are deduced in neuroscience, beyond analog deep neural networks. We have already started such a collaboration with neuroscience, supported by the MIT Quest for Intelligence,” adds Yildiz.

      “The collaboration that we have is going to be essential to innovate in the future. The path forward is still going to be very challenging, but at the same time it is very exciting,” del Alamo says.

      “Intercalation reactions such as those found in lithium-ion batteries have been explored extensively for memory devices. This work demonstrates that proton-based memory devices deliver impressive and surprising switching speed and endurance,” says William Chueh, associate professor of materials science and engineering at Stanford University, who was not involved with this research. “It lays the foundation for a new class of memory devices for powering deep learning algorithms.”

      “This work demonstrates a significant breakthrough in biologically inspired resistive-memory devices. These all-solid-state protonic devices are based on exquisite atomic-scale control of protons, similar to biological synapses but at orders of magnitude faster rates,” says Elizabeth Dickey, the Teddy & Wilton Hawkins Distinguished Professor and head of the Department of Materials Science and Engineering at Carnegie Mellon University, who was not involved with this work. “I commend the interdisciplinary MIT team for this exciting development, which will enable future-generation computational devices.”

      This research is funded, in part, by the MIT-IBM Watson AI Lab. More

    • 125 Shares149 Views

      in Energy

      Explained: Why perovskites could take solar cells to new heights

      by

      Perovskites hold promise for creating solar panels that could be easily deposited onto most surfaces, including flexible and textured ones. These materials would also be lightweight, cheap to produce, and as efficient as today’s leading photovoltaic materials, which are mainly silicon. They’re the subject of increasing research and investment, but companies looking to harness their potential do have to address some remaining hurdles before perovskite-based solar cells can be commercially competitive.

      The term perovskite refers not to a specific material, like silicon or cadmium telluride, other leading contenders in the photovoltaic realm, but to a whole family of compounds. The perovskite family of solar materials is named for its structural similarity to a mineral called perovskite, which was discovered in 1839 and named after Russian mineralogist L.A. Perovski.

      The original mineral perovskite, which is calcium titanium oxide (CaTiO3), has a distinctive crystal configuration. It has a three-part structure, whose components have come to be labeled A, B and X, in which lattices of the different components are interlaced. The family of perovskites consists of the many possible combinations of elements or molecules that can occupy each of the three components and form a structure similar to that of the original perovskite itself. (Some researchers even bend the rules a little by naming other crystal structures with similar elements “perovskites,” although this is frowned upon by crystallographers.)

      “You can mix and match atoms and molecules into the structure, with some limits. For instance, if you try to stuff a molecule that’s too big into the structure, you’ll distort it. Eventually you might cause the 3D crystal to separate into a 2D layered structure, or lose ordered structure entirely,” says Tonio Buonassisi, professor of mechanical engineering at MIT and director of the Photovoltaics Research Laboratory. “Perovskites are highly tunable, like a build-your-own-adventure type of crystal structure,” he says.

      That structure of interlaced lattices consists of ions or charged molecules, two of them (A and B) positively charged and the other one (X) negatively charged. The A and B ions are typically of quite different sizes, with the A being larger. 

      Within the overall category of perovskites, there are a number of types, including metal oxide perovskites, which have found applications in catalysis and in energy storage and conversion, such as in fuel cells and metal-air batteries. But a main focus of research activity for more than a decade has been on lead halide perovskites, according to Buonassisi says.

      Within that category, there is still a legion of possibilities, and labs around the world are racing through the tedious work of trying to find the variations that show the best performance in efficiency, cost, and durability — which has so far been the most challenging of the three.

      Many teams have also focused on variations that eliminate the use of lead, to avoid its environmental impact. Buonassisi notes, however, that “consistently over time, the lead-based devices continue to improve in their performance, and none of the other compositions got close in terms of electronic performance.” Work continues on exploring alternatives, but for now none can compete with the lead halide versions.

      One of the great advantages perovskites offer is their great tolerance of defects in the structure, he says. Unlike silicon, which requires extremely high purity to function well in electronic devices, perovskites can function well even with numerous imperfections and impurities.

      Searching for promising new candidate compositions for perovskites is a bit like looking for a needle in a haystack, but recently researchers have come up with a machine-learning system that can greatly streamline this process. This new approach could lead to a much faster development of new alternatives, says Buonassisi, who was a co-author of that research.

      While perovskites continue to show great promise, and several companies are already gearing up to begin some commercial production, durability remains the biggest obstacle they face. While silicon solar panels retain up to 90 percent of their power output after 25 years, perovskites degrade much faster. Great progress has been made — initial samples lasted only a few hours, then weeks or months, but newer formulations have usable lifetimes of up to a few years, suitable for some applications where longevity is not essential.

      From a research perspective, Buonassisi says, one advantage of perovskites is that they are relatively easy to make in the lab — the chemical constituents assemble readily. But that’s also their downside: “The material goes together very easily at room temperature,” he says, “but it also comes apart very easily at room temperature. Easy come, easy go!”

      To deal with that issue, most researchers are focused on using various kinds of protective materials to encapsulate the perovskite, protecting it from exposure to air and moisture. But others are studying the exact mechanisms that lead to that degradation, in hopes of finding formulations or treatments that are more inherently robust. A key finding is that a process called autocatalysis is largely to blame for the breakdown.

      In autocatalysis, as soon as one part of the material starts to degrade, its reaction products act as catalysts to start degrading the neighboring parts of the structure, and a runaway reaction gets underway. A similar problem existed in the early research on some other electronic materials, such as organic light-emitting diodes (OLEDs), and was eventually solved by adding additional purification steps to the raw materials, so a similar solution may be found in the case of perovskites, Buonassisi suggests.

      Buonassisi and his co-researchers recently completed a study showing that once perovskites reach a usable lifetime of at least a decade, thanks to their much lower initial cost that would be sufficient to make them economically viable as a substitute for silicon in large, utility-scale solar farms.

      Overall, progress in the development of perovskites has been impressive and encouraging, he says. With just a few years of work, it has already achieved efficiencies comparable to levels that cadmium telluride (CdTe), “which has been around for much longer, is still struggling to achieve,” he says. “The ease with which these higher performances are reached in this new material are almost stupefying.” Comparing the amount of research time spent to achieve a 1 percent improvement in efficiency, he says, the progress on perovskites has been somewhere between 100 and 1000 times faster than that on CdTe. “That’s one of the reasons it’s so exciting,” he says. More

    • 125 Shares119 Views

      in Energy

      MIT engineers design surfaces that make water boil more efficiently

      by

      The boiling of water or other fluids is an energy-intensive step at the heart of a wide range of industrial processes, including most electricity generating plants, many chemical production systems, and even cooling systems for electronics.

      Improving the efficiency of systems that heat and evaporate water could significantly reduce their energy use. Now, researchers at MIT have found a way to do just that, with a specially tailored surface treatment for the materials used in these systems.

      The improved efficiency comes from a combination of three different kinds of surface modifications, at different size scales. The new findings are described in the journal Advanced Materials in a paper by recent MIT graduate Youngsup Song PhD ’21, Ford Professor of Engineering Evelyn Wang, and four others at MIT. The researchers note that this initial finding is still at a laboratory scale, and more work is needed to develop a practical, industrial-scale process.

      There are two key parameters that describe the boiling process: the heat transfer coefficient (HTC) and the critical heat flux (CHF). In materials design, there’s generally a tradeoff between the two, so anything that improves one of these parameters tends to make the other worse. But both are important for the efficiency of the system, and now, after years of work, the team has achieved a way of significantly improving both properties at the same time, through their combination of different textures added to a material’s surface.

      “Both parameters are important,” Song says, “but enhancing both parameters together is kind of tricky because they have intrinsic trade off.” The reason for that, he explains, is “because if we have lots of bubbles on the boiling surface, that means boiling is very efficient, but if we have too many bubbles on the surface, they can coalesce together, which can form a vapor film over the boiling surface.” That film introduces resistance to the heat transfer from the hot surface to the water. “If we have vapor in between the surface and water, that prevents the heat transfer efficiency and lowers the CHF value,” he says.

      Song, who is now a postdoc at Lawrence Berkeley National Laboratory, carried out much of the research as part of his doctoral thesis work at MIT. While the various components of the new surface treatment he developed had been previously studied, the researchers say this work is the first to show that these methods could be combined to overcome the tradeoff between the two competing parameters.

      Adding a series of microscale cavities, or dents, to a surface is a way of controlling the way bubbles form on that surface, keeping them effectively pinned to the locations of the dents and preventing them from spreading out into a heat-resisting film. In this work, the researchers created an array of 10-micrometer-wide dents separated by about 2 millimeters to prevent film formation. But that separation also reduces the concentration of bubbles at the surface, which can reduce the boiling efficiency. To compensate for that, the team introduced a much smaller-scale surface treatment, creating tiny bumps and ridges at the nanometer scale, which increases the surface area and promotes the rate of evaporation under the bubbles.

      In these experiments, the cavities were made in the centers of a series of pillars on the material’s surface. These pillars, combined with nanostructures, promote wicking of liquid from the base to their tops, and this enhances the boiling process by providing more surface area exposed to the water. In combination, the three “tiers” of the surface texture — the cavity separation, the posts, and the nanoscale texturing — provide a greatly enhanced efficiency for the boiling process, Song says.

      “Those micro cavities define the position where bubbles come up,” he says. “But by separating those cavities by 2 millimeters, we separate the bubbles and minimize the coalescence of bubbles.” At the same time, the nanostructures promote evaporation under the bubbles, and the capillary action induced by the pillars supplies liquid to the bubble base. That maintains a layer of liquid water between the boiling surface and the bubbles of vapor, which enhances the maximum heat flux.

      Although their work has confirmed that the combination of these kinds of surface treatments can work and achieve the desired effects, this work was done under small-scale laboratory conditions that could not easily be scaled up to practical devices, Wang says. “These kinds of structures we’re making are not meant to be scaled in its current form,” she says, but rather were used to prove that such a system can work. One next step will be to find alternative ways of creating these kinds of surface textures so these methods could more easily be scaled up to practical dimensions.

      “Showing that we can control the surface in this way to get enhancement is a first step,” she says. “Then the next step is to think about more scalable approaches.” For example, though the pillars on the surface in these experiments were created using clean-room methods commonly used to produce semiconductor chips, there are other, less demanding ways of creating such structures, such as electrodeposition. There are also a number of different ways to produce the surface nanostructure textures, some of which may be more easily scalable.

      There may be some significant small-scale applications that could use this process in its present form, such as the thermal management of electronic devices, an area that is becoming more important as semiconductor devices get smaller and managing their heat output becomes ever more important. “There’s definitely a space there where this is really important,” Wang says.

      Even those kinds of applications will take some time to develop because typically thermal management systems for electronics use liquids other than water, known as dielectric liquids. These liquids have different surface tension and other properties than water, so the dimensions of the surface features would have to be adjusted accordingly. Work on these differences is one of the next steps for the ongoing research, Wang says.

      This same multiscale structuring technique could also be applied to different liquids, Song says, by adjusting the dimensions to account for the different properties of the liquids. “Those kinds of details can be changed, and that can be our next step,” he says.

      The team also included Carlos Diaz-Martin, Lenan Zhang, Hyeongyun Cha, and Yajing Zhao, all at MIT. The work was supported by the Advanced Research Projects Agency-Energy (ARPA-E), the Air Force Office of Scientific Research, and the Singapore-MIT Alliance for Research and Technology, and made use of the MIT.nano facilities. More

    • 63 Shares169 Views

      in Security

      MIT engineers introduce the Oreometer

      by

      When you twist open an Oreo cookie to get to the creamy center, you’re mimicking a standard test in rheology — the study of how a non-Newtonian material flows when twisted, pressed, or otherwise stressed. MIT engineers have now subjected the sandwich cookie to rigorous materials tests to get to the center of a tantalizing question: Why does the cookie’s cream stick to just one wafer when twisted apart?

      “There’s the fascinating problem of trying to get the cream to distribute evenly between the two wafers, which turns out to be really hard,” says Max Fan, an undergraduate in MIT’s Department of Mechanical Engineering.

      In pursuit of an answer, the team subjected cookies to standard rheology tests in the lab and found that no matter the flavor or amount of stuffing, the cream at the center of an Oreo almost always sticks to one wafer when twisted open. Only for older boxes of cookies does the cream sometimes separate more evenly between both wafers.

      The researchers also measured the torque required to twist open an Oreo, and found it to be similar to the torque required to turn a doorknob and about 1/10th what’s needed to twist open a bottlecap. The cream’s failure stress — i.e. the force per area required to get the cream to flow, or deform — is twice that of cream cheese and peanut butter, and about the same magnitude as mozzarella cheese. Judging from the cream’s response to stress, the team classifies its texture as “mushy,” rather than brittle, tough, or rubbery.

      So, why does the cookie’s cream glom to one side rather than splitting evenly between both? The manufacturing process may be to blame.

      “Videos of the manufacturing process show that they put the first wafer down, then dispense a ball of cream onto that wafer before putting the second wafer on top,” says Crystal Owens, an MIT mechanical engineering PhD candidate who studies the properties of complex fluids. “Apparently that little time delay may make the cream stick better to the first wafer.”

      The team’s study isn’t simply a sweet diversion from bread-and-butter research; it’s also an opportunity to make the science of rheology accessible to others. To that end, the researchers have designed a 3D-printable “Oreometer” — a simple device that firmly grasps an Oreo cookie and uses pennies and rubber bands to control the twisting force that progressively twists the cookie open. Instructions for the tabletop device can be found here.

      The new study, “On Oreology, the fracture and flow of ‘milk’s favorite cookie,’” appears today in Kitchen Flows, a special issue of the journal Physics of Fluids. It was conceived of early in the Covid-19 pandemic, when many scientists’ labs were closed or difficult to access. In addition to Owens and Fan, co-authors are mechanical engineering professors Gareth McKinley and A. John Hart.

      Confection connection

      A standard test in rheology places a fluid, slurry, or other flowable material onto the base of an instrument known as a rheometer. A parallel plate above the base can be lowered onto the test material. The plate is then twisted as sensors track the applied rotation and torque.

      Owens, who regularly uses a laboratory rheometer to test fluid materials such as 3D-printable inks, couldn’t help noting a similarity with sandwich cookies. As she writes in the new study:

      “Scientifically, sandwich cookies present a paradigmatic model of parallel plate rheometry in which a fluid sample, the cream, is held between two parallel plates, the wafers. When the wafers are counter-rotated, the cream deforms, flows, and ultimately fractures, leading to separation of the cookie into two pieces.”

      While Oreo cream may not appear to possess fluid-like properties, it is considered a “yield stress fluid” — a soft solid when unperturbed that can start to flow under enough stress, the way toothpaste, frosting, certain cosmetics, and concrete do.

      Curious as to whether others had explored the connection between Oreos and rheology, Owens found mention of a 2016 Princeton University study in which physicists first reported that indeed, when twisting Oreos by hand, the cream almost always came off on one wafer.

      “We wanted to build on this to see what actually causes this effect and if we could control it if we mounted the Oreos carefully onto our rheometer,” she says.

      Play video

      Cookie twist

      In an experiment that they would repeat for multiple cookies of various fillings and flavors, the researchers glued an Oreo to both the top and bottom plates of a rheometer and applied varying degrees of torque and angular rotation, noting the values  that successfully twisted each cookie apart. They plugged the measurements into equations to calculate the cream’s viscoelasticity, or flowability. For each experiment, they also noted the cream’s “post-mortem distribution,” or where the cream ended up after twisting open.

      In all, the team went through about 20 boxes of Oreos, including regular, Double Stuf, and Mega Stuf levels of filling, and regular, dark chocolate, and “golden” wafer flavors. Surprisingly, they found that no matter the amount of cream filling or flavor, the cream almost always separated onto one wafer.

      “We had expected an effect based on size,” Owens says. “If there was more cream between layers, it should be easier to deform. But that’s not actually the case.”

      Curiously, when they mapped each cookie’s result to its original position in the box, they noticed the cream tended to stick to the inward-facing wafer: Cookies on the left side of the box twisted such that the cream ended up on the right wafer, whereas cookies on the right side separated with cream mostly on the left wafer. They suspect this box distribution may be a result of post-manufacturing environmental effects, such as heating or jostling that may cause cream to peel slightly away from the outer wafers, even before twisting.

      The understanding gained from the properties of Oreo cream could potentially be applied to the design of other complex fluid materials.

      “My 3D printing fluids are in the same class of materials as Oreo cream,” she says. “So, this new understanding can help me better design ink when I’m trying to print flexible electronics from a slurry of carbon nanotubes, because they deform in almost exactly the same way.”

      As for the cookie itself, she suggests that if the inside of Oreo wafers were more textured, the cream might grip better onto both sides and split more evenly when twisted.

      “As they are now, we found there’s no trick to twisting that would split the cream evenly,” Owens concludes.

      This research was supported, in part, by the MIT UROP program and by the National Defense Science and Engineering Graduate Fellowship Program. More

    • 188 Shares129 Views

      in Energy

      A new heat engine with no moving parts is as efficient as a steam turbine

      by

      Engineers at MIT and the National Renewable Energy Laboratory (NREL) have designed a heat engine with no moving parts. Their new demonstrations show that it converts heat to electricity with over 40 percent efficiency — a performance better than that of traditional steam turbines.

      The heat engine is a thermophotovoltaic (TPV) cell, similar to a solar panel’s photovoltaic cells, that passively captures high-energy photons from a white-hot heat source and converts them into electricity. The team’s design can generate electricity from a heat source of between 1,900 to 2,400 degrees Celsius, or up to about 4,300 degrees Fahrenheit.

      The researchers plan to incorporate the TPV cell into a grid-scale thermal battery. The system would absorb excess energy from renewable sources such as the sun and store that energy in heavily insulated banks of hot graphite. When the energy is needed, such as on overcast days, TPV cells would convert the heat into electricity, and dispatch the energy to a power grid.

      With the new TPV cell, the team has now successfully demonstrated the main parts of the system in separate, small-scale experiments. They are working to integrate the parts to demonstrate a fully operational system. From there, they hope to scale up the system to replace fossil-fuel-driven power plants and enable a fully decarbonized power grid, supplied entirely by renewable energy.

      “Thermophotovoltaic cells were the last key step toward demonstrating that thermal batteries are a viable concept,” says Asegun Henry, the Robert N. Noyce Career Development Professor in MIT’s Department of Mechanical Engineering. “This is an absolutely critical step on the path to proliferate renewable energy and get to a fully decarbonized grid.”

      Henry and his collaborators have published their results today in the journal Nature. Co-authors at MIT include Alina LaPotin, Kevin Schulte, Kyle Buznitsky, Colin Kelsall, Andrew Rohskopf, and Evelyn Wang, the Ford Professor of Engineering and head of the Department of Mechanical Engineering, along with collaborators at NREL in Golden, Colorado.

      Jumping the gap

      More than 90 percent of the world’s electricity comes from sources of heat such as coal, natural gas, nuclear energy, and concentrated solar energy. For a century, steam turbines have been the industrial standard for converting such heat sources into electricity.

      On average, steam turbines reliably convert about 35 percent of a heat source into electricity, with about 60 percent representing the highest efficiency of any heat engine to date. But the machinery depends on moving parts that are temperature- limited. Heat sources higher than 2,000 degrees Celsius, such as Henry’s proposed thermal battery system, would be too hot for turbines.

      In recent years, scientists have looked into solid-state alternatives — heat engines with no moving parts, that could potentially work efficiently at higher temperatures.

      “One of the advantages of solid-state energy converters are that they can operate at higher temperatures with lower maintenance costs because they have no moving parts,” Henry says. “They just sit there and reliably generate electricity.”

      Thermophotovoltaic cells offered one exploratory route toward solid-state heat engines. Much like solar cells, TPV cells could be made from semiconducting materials with a particular bandgap — the gap between a material’s valence band and its conduction band. If a photon with a high enough energy is absorbed by the material, it can kick an electron across the bandgap, where the electron can then conduct, and thereby generate electricity — doing so without moving rotors or blades.

      To date, most TPV cells have only reached efficiencies of around 20 percent, with the record at 32 percent, as they have been made of relatively low-bandgap materials that convert lower-temperature, low-energy photons, and therefore convert energy less efficiently.

      Catching light

      In their new TPV design, Henry and his colleagues looked to capture higher-energy photons from a higher-temperature heat source, thereby converting energy more efficiently. The team’s new cell does so with higher-bandgap materials and multiple junctions, or material layers, compared with existing TPV designs.

      The cell is fabricated from three main regions: a high-bandgap alloy, which sits over a slightly lower-bandgap alloy, underneath which is a mirror-like layer of gold. The first layer captures a heat source’s highest-energy photons and converts them into electricity, while lower-energy photons that pass through the first layer are captured by the second and converted to add to the generated voltage. Any photons that pass through this second layer are then reflected by the mirror, back to the heat source, rather than being absorbed as wasted heat.

      The team tested the cell’s efficiency by placing it over a heat flux sensor — a device that directly measures the heat absorbed from the cell. They exposed the cell to a high-temperature lamp and concentrated the light onto the cell. They then varied the bulb’s intensity, or temperature, and observed how the cell’s power efficiency — the amount of power it produced, compared with the heat it absorbed — changed with temperature. Over a range of 1,900 to 2,400 degrees Celsius, the new TPV cell maintained an efficiency of around 40 percent.

      “We can get a high efficiency over a broad range of temperatures relevant for thermal batteries,” Henry says.

      The cell in the experiments is about a square centimeter. For a grid-scale thermal battery system, Henry envisions the TPV cells would have to scale up to about 10,000 square feet (about a quarter of a football field), and would operate in climate-controlled warehouses to draw power from huge banks of stored solar energy. He points out that an infrastructure exists for making large-scale photovoltaic cells, which could also be adapted to manufacture TPVs.

      “There’s definitely a huge net positive here in terms of sustainability,” Henry says. “The technology is safe, environmentally benign in its life cycle, and can have a tremendous impact on abating carbon dioxide emissions from electricity production.”

      This research was supported, in part, by the U.S. Department of Energy. More

    • 138 Shares99 Views

      in Energy

      Engineers enlist AI to help scale up advanced solar cell manufacturing

      by

      Perovskites are a family of materials that are currently the leading contender to potentially replace today’s silicon-based solar photovoltaics. They hold the promise of panels that are far thinner and lighter, that could be made with ultra-high throughput at room temperature instead of at hundreds of degrees, and that are cheaper and easier to transport and install. But bringing these materials from controlled laboratory experiments into a product that can be manufactured competitively has been a long struggle.

      Manufacturing perovskite-based solar cells involves optimizing at least a dozen or so variables at once, even within one particular manufacturing approach among many possibilities. But a new system based on a novel approach to machine learning could speed up the development of optimized production methods and help make the next generation of solar power a reality.

      The system, developed by researchers at MIT and Stanford University over the last few years, makes it possible to integrate data from prior experiments, and information based on personal observations by experienced workers, into the machine learning process. This makes the outcomes more accurate and has already led to the manufacturing of perovskite cells with an energy conversion efficiency of 18.5 percent, a competitive level for today’s market.

      The research is reported today in the journal Joule, in a paper by MIT professor of mechanical engineering Tonio Buonassisi, Stanford professor of materials science and engineering Reinhold Dauskardt, recent MIT research assistant Zhe Liu, Stanford doctoral graduate Nicholas Rolston, and three others.

      Perovskites are a group of layered crystalline compounds defined by the configuration of the atoms in their crystal lattice. There are thousands of such possible compounds and many different ways of making them. While most lab-scale development of perovskite materials uses a spin-coating technique, that’s not practical for larger-scale manufacturing, so companies and labs around the world have been searching for ways of translating these lab materials into a practical, manufacturable product.

      “There’s always a big challenge when you’re trying to take a lab-scale process and then transfer it to something like a startup or a manufacturing line,” says Rolston, who is now an assistant professor at Arizona State University. The team looked at a process that they felt had the greatest potential, a method called rapid spray plasma processing, or RSPP.

      The manufacturing process would involve a moving roll-to-roll surface, or series of sheets, on which the precursor solutions for the perovskite compound would be sprayed or ink-jetted as the sheet rolled by. The material would then move on to a curing stage, providing a rapid and continuous output “with throughputs that are higher than for any other photovoltaic technology,” Rolston says.

      “The real breakthrough with this platform is that it would allow us to scale in a way that no other material has allowed us to do,” he adds. “Even materials like silicon require a much longer timeframe because of the processing that’s done. Whereas you can think of [this approach as more] like spray painting.”

      Within that process, at least a dozen variables may affect the outcome, some of them more controllable than others. These include the composition of the starting materials, the temperature, the humidity, the speed of the processing path, the distance of the nozzle used to spray the material onto a substrate, and the methods of curing the material. Many of these factors can interact with each other, and if the process is in open air, then humidity, for example, may be uncontrolled. Evaluating all possible combinations of these variables through experimentation is impossible, so machine learning was needed to help guide the experimental process.

      But while most machine-learning systems use raw data such as measurements of the electrical and other properties of test samples, they don’t typically incorporate human experience such as qualitative observations made by the experimenters of the visual and other properties of the test samples, or information from other experiments reported by other researchers. So, the team found a way to incorporate such outside information into the machine learning model, using a probability factor based on a mathematical technique called Bayesian Optimization.

      Using the system, he says, “having a model that comes from experimental data, we can find out trends that we weren’t able to see before.” For example, they initially had trouble adjusting for uncontrolled variations in humidity in their ambient setting. But the model showed them “that we could overcome our humidity challenges by changing the temperature, for instance, and by changing some of the other knobs.”

      The system now allows experimenters to much more rapidly guide their process in order to optimize it for a given set of conditions or required outcomes. In their experiments, the team focused on optimizing the power output, but the system could also be used to simultaneously incorporate other criteria, such as cost and durability — something members of the team are continuing to work on, Buonassisi says.

      The researchers were encouraged by the Department of Energy, which sponsored the work, to commercialize the technology, and they’re currently focusing on tech transfer to existing perovskite manufacturers. “We are reaching out to companies now,” Buonassisi says, and the code they developed has been made freely available through an open-source server. “It’s now on GitHub, anyone can download it, anyone can run it,” he says. “We’re happy to help companies get started in using our code.”

      Already, several companies are gearing up to produce perovskite-based solar panels, even though they are still working out the details of how to produce them, says Liu, who is now at the Northwestern Polytechnical University in Xi’an, China. He says companies there are not yet doing large-scale manufacturing, but instead starting with smaller, high-value applications such as building-integrated solar tiles where appearance is important. Three of these companies “are on track or are being pushed by investors to manufacture 1 meter by 2-meter rectangular modules [comparable to today’s most common solar panels], within two years,” he says.

      ‘The problem is, they don’t have a consensus on what manufacturing technology to use,” Liu says. The RSPP method, developed at Stanford, “still has a good chance” to be competitive, he says. And the machine learning system the team developed could prove to be important in guiding the optimization of whatever process ends up being used.

      “The primary goal was to accelerate the process, so it required less time, less experiments, and less human hours to develop something that is usable right away, for free, for industry,” he says.

      “Existing work on machine-learning-driven perovskite PV fabrication largely focuses on spin-coating, a lab-scale technique,” says Ted Sargent, University Professor at the University of Toronto, who was not associated with this work, which he says demonstrates “a workflow that is readily adapted to the deposition techniques that dominate the thin-film industry. Only a handful of groups have the simultaneous expertise in engineering and computation to drive such advances.” Sargent adds that this approach “could be an exciting advance for the manufacture of a broader family of materials” including LEDs, other PV technologies, and graphene, “in short, any industry that uses some form of vapor or vacuum deposition.” 

      The team also included Austin Flick and Thomas Colburn at Stanford and Zekun Ren at the Singapore-MIT Alliance for Science and Technology (SMART). In addition to the Department of Energy, the work was supported by a fellowship from the MIT Energy Initiative, the Graduate Research Fellowship Program from the National Science Foundation, and the SMART program. More

    • 75 Shares119 Views

      in Energy

      Chemical reactions for the energy transition

      by

      One challenge in decarbonizing the energy system is knowing how to deal with new types of fuels. Traditional fuels such as natural gas and oil can be combined with other materials and then heated to high temperatures so they chemically react to produce other useful fuels or substances, or even energy to do work. But new materials such as biofuels can’t take as much heat without breaking down.

      A key ingredient in such chemical reactions is a specially designed solid catalyst that is added to encourage the reaction to happen but isn’t itself consumed in the process. With traditional materials, the solid catalyst typically interacts with a gas; but with fuels derived from biomass, for example, the catalyst must work with a liquid — a special challenge for those who design catalysts.

      For nearly a decade, Yogesh Surendranath, an associate professor of chemistry at MIT, has been focusing on chemical reactions between solid catalysts and liquids, but in a different situation: rather than using heat to drive reactions, he and his team input electricity from a battery or a renewable source such as wind or solar to give chemically inactive molecules more energy so they react. And key to their research is designing and fabricating solid catalysts that work well for reactions involving liquids.

      Recognizing the need to use biomass to develop sustainable liquid fuels, Surendranath wondered whether he and his team could take the principles they have learned about designing catalysts to drive liquid-solid reactions with electricity and apply them to reactions that occur at liquid-solid interfaces without any input of electricity.

      To their surprise, they found that their knowledge is directly relevant. Why? “What we found — amazingly — is that even when you don’t hook up wires to your catalyst, there are tiny internal ‘wires’ that do the reaction,” says Surendranath. “So, reactions that people generally think operate without any flow of current actually do involve electrons shuttling from one place to another.” And that means that Surendranath and his team can bring the powerful techniques of electrochemistry to bear on the problem of designing catalysts for sustainable fuels.

      A novel hypothesis

      Their work has focused on a class of chemical reactions important in the energy transition that involve adding oxygen to small organic (carbon-containing) molecules such as ethanol, methanol, and formic acid. The conventional assumption is that the reactant and oxygen chemically react to form the product plus water. And a solid catalyst — often a combination of metals — is present to provide sites on which the reactant and oxygen can interact.

      But Surendranath proposed a different view of what’s going on. In the usual setup, two catalysts, each one composed of many nanoparticles, are mounted on a conductive carbon substrate and submerged in water. In that arrangement, negatively charged electrons can flow easily through the carbon, while positively charged protons can flow easily through water.

      Surendranath’s hypothesis was that the conversion of reactant to product progresses by means of two separate “half-reactions” on the two catalysts. On one catalyst, the reactant turns into a product, in the process sending electrons into the carbon substrate and protons into the water. Those electrons and protons are picked up by the other catalyst, where they drive the oxygen-to-water conversion. So, instead of a single reaction, two separate but coordinated half-reactions together achieve the net conversion of reactant to product.

      As a result, the overall reaction doesn’t actually involve any net electron production or consumption. It is a standard “thermal” reaction resulting from the energy in the molecules and maybe some added heat. The conventional approach to designing a catalyst for such a reaction would focus on increasing the rate of that reactant-to-product conversion. And the best catalyst for that kind of reaction could turn out to be, say, gold or palladium or some other expensive precious metal.

      However, if that reaction actually involves two half-reactions, as Surendranath proposed, there is a flow of electrical charge (the electrons and protons) between them. So Surendranath and others in the field could instead use techniques of electrochemistry to design not a single catalyst for the overall reaction but rather two separate catalysts — one to speed up one half-reaction and one to speed up the other half-reaction. “That means we don’t have to design one catalyst to do all the heavy lifting of speeding up the entire reaction,” says Surendranath. “We might be able to pair up two low-cost, earth-abundant catalysts, each of which does half of the reaction well, and together they carry out the overall transformation quickly and efficiently.”

      But there’s one more consideration: Electrons can flow through the entire catalyst composite, which encompasses the catalyst particle(s) and the carbon substrate. For the chemical conversion to happen as quickly as possible, the rate at which electrons are put into the catalyst composite must exactly match the rate at which they are taken out. Focusing on just the electrons, if the reaction-to-product conversion on the first catalyst sends the same number of electrons per second into the “bath of electrons” in the catalyst composite as the oxygen-to-water conversion on the second catalyst takes out, the two half-reactions will be balanced, and the electron flow — and the rate of the combined reaction — will be fast. The trick is to find good catalysts for each of the half-reactions that are perfectly matched in terms of electrons in and electrons out.

      “A good catalyst or pair of catalysts can maintain an electrical potential — essentially a voltage — at which both half-reactions are fast and are balanced,” says Jaeyune Ryu PhD ’21, a former member of the Surendranath lab and lead author of the study; Ryu is now a postdoc at Harvard University. “The rates of the reactions are equal, and the voltage in the catalyst composite won’t change during the overall thermal reaction.”

      Drawing on electrochemistry

      Based on their new understanding, Surendranath, Ryu, and their colleagues turned to electrochemistry techniques to identify a good catalyst for each half-reaction that would also pair up to work well together. Their analytical framework for guiding catalyst development for systems that combine two half-reactions is based on a theory that has been used to understand corrosion for almost 100 years, but has rarely been applied to understand or design catalysts for reactions involving small molecules important for the energy transition.

      Key to their work is a potentiostat, a type of voltmeter that can either passively measure the voltage of a system or actively change the voltage to cause a reaction to occur. In their experiments, Surendranath and his team use the potentiostat to measure the voltage of the catalyst in real time, monitoring how it changes millisecond to millisecond. They then correlate those voltage measurements with simultaneous but separate measurements of the overall rate of catalysis to understand the reaction pathway.

      For their study of the conversion of small, energy-related molecules, they first tested a series of catalysts to find good ones for each half-reaction — one to convert the reactant to product, producing electrons and protons, and another to convert the oxygen to water, consuming electrons and protons. In each case, a promising candidate would yield a rapid reaction — that is, a fast flow of electrons and protons out or in.

      To help identify an effective catalyst for performing the first half-reaction, the researchers used their potentiostat to input carefully controlled voltages and measured the resulting current that flowed through the catalyst. A good catalyst will generate lots of current for little applied voltage; a poor catalyst will require high applied voltage to get the same amount of current. The team then followed the same procedure to identify a good catalyst for the second half-reaction.

      To expedite the overall reaction, the researchers needed to find two catalysts that matched well — where the amount of current at a given applied voltage was high for each of them, ensuring that as one produced a rapid flow of electrons and protons, the other one consumed them at the same rate.

      To test promising pairs, the researchers used the potentiostat to measure the voltage of the catalyst composite during net catalysis — not changing the voltage as before, but now just measuring it from tiny samples. In each test, the voltage will naturally settle at a certain level, and the goal is for that to happen when the rate of both reactions is high.

      Validating their hypothesis and looking ahead

      By testing the two half-reactions, the researchers could measure how the reaction rate for each one varied with changes in the applied voltage. From those measurements, they could predict the voltage at which the full reaction would proceed fastest. Measurements of the full reaction matched their predictions, supporting their hypothesis.

      The team’s novel approach of using electrochemistry techniques to examine reactions thought to be strictly thermal in nature provides new insights into the detailed steps by which those reactions occur and therefore into how to design catalysts to speed them up. “We can now use a divide-and-conquer strategy,” says Ryu. “We know that the net thermal reaction in our study happens through two ‘hidden’ but coupled half-reactions, so we can aim to optimize one half-reaction at a time” — possibly using low-cost catalyst materials for one or both.

      Adds Surendranath, “One of the things that we’re excited about in this study is that the result is not final in and of itself. It has really seeded a brand-new thrust area in our research program, including new ways to design catalysts for the production and transformation of renewable fuels and chemicals.”

      This research was supported primarily by the Air Force Office of Scientific Research. Jaeyune Ryu PhD ’21 was supported by a Samsung Scholarship. Additional support was provided by a National Science Foundation Graduate Research Fellowship.

      This article appears in the Autumn 2021 issue of Energy Futures, the magazine of the MIT Energy Initiative. More

    • 113 Shares189 Views

      in Energy

      New program bolsters innovation in next-generation artificial intelligence hardware

      by

      The MIT AI Hardware Program is a new academia and industry collaboration aimed at defining and developing translational technologies in hardware and software for the AI and quantum age. A collaboration between the MIT School of Engineering and MIT Schwarzman College of Computing, involving the Microsystems Technologies Laboratories and programs and units in the college, the cross-disciplinary effort aims to innovate technologies that will deliver enhanced energy efficiency systems for cloud and edge computing.

      “A sharp focus on AI hardware manufacturing, research, and design is critical to meet the demands of the world’s evolving devices, architectures, and systems,” says Anantha Chandrakasan, dean of the MIT School of Engineering and Vannevar Bush Professor of Electrical Engineering and Computer Science. “Knowledge-sharing between industry and academia is imperative to the future of high-performance computing.”

      Based on use-inspired research involving materials, devices, circuits, algorithms, and software, the MIT AI Hardware Program convenes researchers from MIT and industry to facilitate the transition of fundamental knowledge to real-world technological solutions. The program spans materials and devices, as well as architecture and algorithms enabling energy-efficient and sustainable high-performance computing.

      “As AI systems become more sophisticated, new solutions are sorely needed to enable more advanced applications and deliver greater performance,” says Daniel Huttenlocher, dean of the MIT Schwarzman College of Computing and Henry Ellis Warren Professor of Electrical Engineering and Computer Science. “Our aim is to devise real-world technological solutions and lead the development of technologies for AI in hardware and software.”

      The inaugural members of the program are companies from a wide range of industries including chip-making, semiconductor manufacturing equipment, AI and computing services, and information systems R&D organizations. The companies represent a diverse ecosystem, both nationally and internationally, and will work with MIT faculty and students to help shape a vibrant future for our planet through cutting-edge AI hardware research.

      The five inaugural members of the MIT AI Hardware Program are:  

      Amazon, a global technology company whose hardware inventions include the Kindle, Amazon Echo, Fire TV, and Astro; 
      Analog Devices, a global leader in the design and manufacturing of analog, mixed signal, and DSP integrated circuits; 
      ASML, an innovation leader in the semiconductor industry, providing chipmakers with hardware, software, and services to mass produce patterns on silicon through lithography; 
      NTT Research, a subsidiary of NTT that conducts fundamental research to upgrade reality in game-changing ways that improve lives and brighten our global future; and 
      TSMC, the world’s leading dedicated semiconductor foundry.

      The MIT AI Hardware Program will create a roadmap of transformative AI hardware technologies. Leveraging MIT.nano, the most advanced university nanofabrication facility anywhere, the program will foster a unique environment for AI hardware research.  

      “We are all in awe at the seemingly superhuman capabilities of today’s AI systems. But this comes at a rapidly increasing and unsustainable energy cost,” says Jesús del Alamo, the Donner Professor in MIT’s Department of Electrical Engineering and Computer Science. “Continued progress in AI will require new and vastly more energy-efficient systems. This, in turn, will demand innovations across the entire abstraction stack, from materials and devices to systems and software. The program is in a unique position to contribute to this quest.”

      The program will prioritize the following topics:

      analog neural networks;
      new roadmap CMOS designs;
      heterogeneous integration for AI systems;
      onolithic-3D AI systems;
      analog nonvolatile memory devices;
      software-hardware co-design;
      intelligence at the edge;
      intelligent sensors;
      energy-efficient AI;
      intelligent internet of things (IIoT);
      neuromorphic computing;
      AI edge security;
      quantum AI;
      wireless technologies;
      hybrid-cloud computing; and
      high-performance computation.

      “We live in an era where paradigm-shifting discoveries in hardware, systems communications, and computing have become mandatory to find sustainable solutions — solutions that we are proud to give to the world and generations to come,” says Aude Oliva, senior research scientist in the MIT Computer Science and Artificial Intelligence Laboratory (CSAIL) and director of strategic industry engagement in the MIT Schwarzman College of Computing.

      The new program is co-led by Jesús del Alamo and Aude Oliva, and Anantha Chandrakasan serves as chair. More