Toyota Research Institute embraces machine learning to advance the switch from internal combustion engine to electric vehicles.
|Dr. Brian Storey gives the keynote address at the MIT MRL Materials Day Symposium Oct. 9, 2019. Storey directs Toyota Research Institute’s accelerated materials design initiative from its Kendall Square office in Cambridge, Mass., and is embracing machine learning to advance the switch from internal combustion engine to electric vehicles. Image, Denis Paiste, Materials Research Laboratory.|
With 100 million Toyota vehicles on the planet emitting greenhouse gases at a rate roughly comparable to those of France, Toyota has set a goal of reducing all tailpipe emissions by 90 percent by the year 2050, according to Dr. Brian Storey, who directs Toyota Research Institute’s Accelerated Materials Design & Discovery program from its Kendall Square office in Cambridge, Mass. He gave the keynote address at the MIT MRL Materials Day Symposium Oct. 9, 2019.
“A rapid shift from the traditional vehicle to electric vehicles has started,” Storey says. “And we want to enable that to happen at a faster pace.”
“Our role at TRI is to develop tools for accelerating the development of emissions free vehicles,” Storey says. He says machine learning is helping to speed up those innovations, but the challenges are very great, so his team has to be a little humble about what it can actually accomplish.
Electrification is just one of four “disrupters” to the automotive industry, that are often abbreviated CASE (Connected, Autonomous, Shared, Electric). “It’s a disrupter to the industry because Toyota has decades of experience of optimizing the combustion engine,” Storey says. “We know how to do it; it’s reliable; it’s affordable; it lasts forever. Really the heart of the Toyota brand is the quality of the combustion engine and transmission.”
Storey states that as society shifts toward electrification – battery or fuel cell vehicles – new capability, technology and know-how is needed. Storey says “while Toyota has a lot of experience in these areas, we still need to move faster if we are going to make this kind of transition.”
To help with that acceleration, Toyota Research Institute is providing $10 million a year to support research of approximately 125 professors, postdocs and graduate students at 10 academic institutions. About $2 million a year of that research is being done at MIT. Storey is also a Professor of Mechanical Engineering at Olin College of Engineering.
For example, the Battery Evaluation and Early Prediction (BEEP) project, which is a TRI collaboration with MIT and Stanford, aims to expand the value of lithium-based battery systems. In experiments, many batteries are charged and discharged at the same time. “From that data alone, the charge and discharge data, we can extract features. It’s super practical because we get the data. We extract features from the data, and we can correlate those features with lifetime,” Storey explains.
The traditional way of testing whether a battery is going to last for a thousand cycles, is to cycle it for a thousand times. Storey notes that if each cycle takes one hour, one battery requires 1,000 hours of testing. “What we want to do is bring that time way back, and so our goal is to able to do it in 5, to cycle 5 times, and get a good estimate of what the battery’s lifetime would be at 1,000 cycles doing it purely from data,” Storey says.
“Our dream, which is a work in progress, is to have a system architecture that overlies all these projects and can start to tie them together. We are creating a system that’s built for machine learning from the start…”
– Brian Storey, Director of Accelerated Materials Design & Discovery at Toyota Research Institute (TRI)
Published results in Nature Energy in March 2019 show just a 4.9% test error using data in classifying lithium-ion batteries from the first five charge/discharge cycles.
“This is a nice capability because it actually allows acceleration in testing,” Storey notes. “It’s using machine learning, but it’s really using it at the device scale, the ‘as-manufactured’ battery.”
The cloud-based battery evaluation software system allows TRI to collaborate easily with colleagues at MIT, Stanford and Toyota’s home base in Japan, he says.
Program researchers operate it in a closed loop, semi-autonomous way, where the computer decides and executes the next best experiment. The system finds charging policies that are better than ones that have been published in the literature, and it finds them rapidly. “The key to this is the early prediction model, because if we want to predict the lifetime, we don’t have to do the whole test.” Storey adds that the closed loop testing “pulls the scientist up a level in terms of what questions they can ask.”
TRI would like to use this closed loop battery evaluation system to optimize the first charge/discharge cycle a battery goes through, which is called formation cycling. “It’s like caring for the battery when it’s a baby,” Storey explains. “How you do those first cycles, actually sets it up for the rest of its life. It’s a real black art and how do you optimize this process?”
TRI’s long-term term goal is to improve battery durability so that, from the consumer point of view, the battery capacity never goes down. Storey emphasizes “we want the battery in the car to just last forever.”
Storey notes TRI is also conducting two other research projects, AI-Assisted Catalysis Experimentation (ACE) with CalTech to improve catalysts for fuel cell vehicles such as Toyota’s Mirai, and a materials synthesis project, mostly within TRI, to use machine-learning to identify whether or not the new materials predicted on the computer are likely to be synthesizable.
For the materials synthesis project, TRI began with the phase diagrams of materials. “You build up a network of every material you’ve got in the computational database and look at features of the network. Believing that somehow those materials are connected to other materials through the relationship in this network, provides a prediction of synthesizability,” Storey explains. “The way you can train the algorithm is by looking in the historical record of when certain materials were synthesized. You can virtually roll the clock back, pretending to know only what you knew in 1980, and use that to train your algorithm.” A report on the materials synthesis network was published in May 2019 in Nature Communications.
|Dr. Brian Storey, Director of Accelerated Materials Design & Discovery at Toyota Research Institute (TRI), speaks at the MIT MRL Materials Day Symposium Oct. 9, 2019. Image, Denis Paiste, Materials Research Laboratory.|
TRI is collaborating with Lawrence Berkeley National Laboratory and MIT Professor Martin Z. Bazant on a project that couples highly detailed mechanics of battery particles revealed through 4D scanning tunneling electron microscopy with a continuum model that captures larger scale materials properties. “This program figures out the reaction kinetics and thermodynamics at a continuum scale which is otherwise unknown,” Storey says.
“We’re putting our software tools online, so over the coming year, many of these tools will start becoming available,” Storey says. Hosted by Lawrence Berkeley, the Propnet materials database is already accessible to internal collaborators. Matscholar is accessible through GitHub. Both projects were funded by TRI.
“Our dream, which is a work in progress, is to have a system architecture that overlies all these projects and can start to tie them together. We are creating a system that’s built for machine learning from the start, allows for diverse data, allows for systems and atom scale measurements, and is capable of this idea of AI-driven feedback and autonomy. The idea is that you launch the system and it runs on its own, and everything lives in the cloud to enable collaboration,” Storey says.
Ability to predict and make new materials faster highlights need for safety, reliability and accurate data.
The promise and challenges of artificial intelligence and machine learning highlighted the MIT Materials Day Symposium Oct. 9, 2019, with presentations on new ways of forming zeolites, faster drug synthesis, advanced optical devices, and more.
“Machine learning is having an impact in all areas of materials research,” Materials Research Laboratory Director Carl V. Thompson says.
“We’re increasingly able to work in tandem with machines to help us decide what materials to make,” says Elsa A. Olivetti, the Atlantic Richfield Associate Professor of Energy Studies. Machine learning is also guiding how to make those materials with new insights into synthesis methods, and in some cases, such as with robotic systems, actually making those materials, she notes.
Keynote speaker Brian Storey, Director of Accelerated Materials Design & Discovery at Toyota Research Institute, spoke about machine learning to advance the switch from internal combustion engine to electric vehicles, and Professor Ju Li, Battelle Energy Alliance Professor of Nuclear Science and Engineering, and Professor of Materials Science and Engineering, spoke about atomic engineering using elastic strain and radiation nudging of atoms.
Olivetti and Rafael Gomez-Bombarelli, the Toyota Assistant Professor in Materials Processing, worked together to apply machine learning to develop a better understanding of porous materials called zeolites, formed from silicon and aluminum oxide, that have a wide range of uses from cat litter to petroleum refining.
“Essentially the idea is that the pore has the right size to hold organic molecules,” Gomez-Bombarelli says. While only about 250 zeolites of this class are known to engineers, physicists can calculate hundreds of thousands of possible ways these structures can form. “Some of them can be converted into each other,” he says. “So, you could mine one zeolite, put it under pressure, or heat it up, and it becomes a different one that could be more valuable for a specific application.”
A traditional method was to interpret these crystalline structures as a combination of building blocks. However, when zeolite transformations were analyzed, more than half the time there were no building blocks in common between the original zeolite before the change and the new zeolite after the change. “Building block theory has some interesting ingredients, but doesn’t quite explain the rules to go from A to B,” Gomez-Bombarelli says.
Gomez-Bombarelli’s new graph-based approach finds that when each zeolite framework structure is represented as a graph, these graphs match before and after in zeolite transformation pairs. “Some classes of transformations only happen between zeolites that have the same graph,” he says.
This work evolved from Olivetti’s data mining of 2.5 million materials science journal articles to uncover recipes for making different inorganic materials. The zeolite study examined 70,000 papers. “One of the challenges in learning from the literature is we publish positive examples, we publish data of things that went well,” Olivetti says. In the zeolite community, researchers also publish what doesn’t work. “That’s a valuable dataset for us to learn from,” she says. “What we’ve been able to use this data set for is to try to predict potential synthesis pathways for making particular types of zeolites.”
In earlier work with colleagues at UMass, Olivetti developed a system that identified common scientific words and techniques found in sentences across this large library and brought together similar findings. “One important challenge in natural language processing is to draw this linked information across a document,” Olivetti explains. “We are trying to build tools that are able to do that linking,” Olivetti says.
AI-assisted chemical synthesis
Klavs F. Jensen, the Warren K. Lewis Professor of Chemical Engineering and Professor of Materials Science and Engineering, described a chemical synthesis system that combines artificial intelligence-guided processing steps with a robotically operated modular reaction system.
For those unfamiliar with synthesis, Jensen explains that “You have reactants you start with, you have reagents that you have to add, catalysts and so forth to make the reaction go, you have intermediates, and ultimately you end up with your product.”
The artificial intelligence system combed 12.5 million reactions, creating a set of rules, or library, from about 160,000 of the most commonly used synthesis recipes, Jensen relates. This machine learning approach suggests processing conditions such as what catalysts, solvents and reagents to use in the reaction.
“You can have the system take whatever information it got from the published literature about conditions and so on and you can use that to form a recipe,” he says. Because there is not enough data yet to inform the system, a chemical expert still needs to step in to specify concentrations, flow rates, and process stack configurations and to ensure safety before sending the recipe to the robotic system.
The researchers demonstrated this system by predicting synthesis plans for 15 drugs or drug-like molecules, for example, the pain killer lidocaine and several high blood pressure drugs, and then making them with the system. The flow reactor system contrasts with a batch system. “In order to be able to accelerate the reactions, we use typically much more aggressive conditions than are done in batch – high temperatures and higher pressures,” Jensen says.
The modular system consists of a processing tower with interchangeable reaction modules and a set of different reagents, which are connected together by the robot for each synthesis. These findings were reported in Science.
Former PhD students Connor W. Coley and Dale A. Thomas built the computer-aided synthesis planner and the flow reactor system, respectively, and former postdoc Justin A. M. Lummiss did the chemistry along with a large team of MIT UROPs, PhD students, and postdocs. Jensen also notes contributions from MIT faculty colleagues Regina Barzilay, William H. Green, A. John Hart, Tommi Jaakkola and Tim Jamison. MIT has filed a patent for the robotic handling of fluid connections. The software suite that suggests and prioritizes possible synthesis routes is open-source and an online version is at the web address, ASKCOS.mit.edu.
Robustness in Machine Learning
Deep learning systems perform amazingly well on benchmark tasks such as images and natural language processing applications, says Professor Asu Ozdaglar, who heads MIT’s Department of Electrical Engineering & Computer Science. Still, researchers are far from understanding why these deep learning systems work, when they will work, and how they generalize. And when they get things wrong, they can go completely awry.
Ozdaglar gave an example of an image with a state-of-the-art classifier that can look at a picture of a cute pig and recognize the image as that of a pig. But, “If you add a little bit of, very little, perturbation, what happens is basically the same classifier thinks that’s an airliner,” Ozdaglar says. “So this is sort of an example where people say machine learning is so powerful, it can make pigs fly,” she says, accompanied by audience laughter. “And this immediately tells us basically we have to go beyond our standard approaches.”
A potential solution lies in an optimization formulation known as a Minimax, or MinMax, problem. Another place where MinMax formulation arises is in Generative Adversarial Network, or GAN, training. Using an example of images of real cars and fake images of cars, Ozdaglar explains, “We would like these fake images to be drawn from the same distribution as the training set, and this is achieved using two neural networks competing with each other, a generator network and a discriminator network. The generator network creates from random noise these fake images that the discriminator network tries to pull apart to see whether this is real or fake.”
“It’s basically another MinMax problem whereby the generator is trying to minimize the distance between these two distributions, fake and real. And then the discriminator is trying to maximize that,” she says. The MinMax problem approach has become the backbone of robust training of deep learning systems, she notes.
Ozdaglar notes that EECS faculty are applying machine learning to new areas including health care, citing the work of Regina Barzilay in detecting breast cancer and David Sontag in using electronic medical records for medical diagnosis and treatment.
The EECS undergraduate machine learning course (6.036) hosted 800 students last spring, and consistently has 600 or more students enrolled, making it the most popular course at MIT. The new Stephen A. Schwarzman College of Computing provides an opportunity to create a more dynamic and adaptable structure than MIT’s traditional department structure. For example, one idea is to create several cross-departmental teaching groups. “We envision things like courses in the foundations of computing, computational science and engineering, social studies of computing, and have these courses taken by all of our students taught jointly by our faculty across MIT,” she says.
Juejun (JJ) Hu, Associate Professor of Materials Science and Engineering, detailed his research coupling a silicon chip-based spectrometer for detecting infrared light wavelengths to a newly created machine learning algorithm. Ordinary spectrometers, going back to Isaac Newton’s first prism, work by splitting light, which reduces intensity but Hu’s version collects all of the light at a single detector which preserves light intensity, but then poses the problem of identifying different wavelengths from a single capture.
“If you want to solve this trade-off between the (spectral) resolution and the signal-to-noise ratio what you have to do is resort to a new type of spectroscopy tool called wavelength multiplexing spectrometer,” Hu says. His new spectrometer architecture, which is called digital Fourier transform spectroscopy, incorporates tunable optical switches on a silicon chip. The device works by measuring the intensity of light at different optical switch settings and comparing the results. “What you have is essentially a group of linear equations that gives you some linear combination of the light intensity at different wavelengths in the form of a detector reading,” he says.
A prototype device with six switches supports a total of 64 unique optical states, which can provide 64 independent readings. “The advantage of this new device architecture is that the performance doubles every time you add a new switch,” he says. Working with Brando Miranda at the Center for Brains Minds and Machines at MIT, he developed a new algorithm, Elastic D1, that gives a resolution down to 0.2 nanometers and gives an accurate light measurement with only two consecutive measurements.
“We believe this kind of unique combination between the hardware of a new spectrometer architecture and the algorithm can enable a wide range of applications ranging from industrial process monitoring to medical imaging,” Hu says. Hu also is applying machine learning in his work on complex optical media such as metasurfaces, which are new optical devices featuring an array of specially designed optical antennas that add a phase delay to the incoming light.
Poster Session Winners
Nineteen MIT Postdocs and graduate students gave two-minute talks about their research during a Poster Session Preview. At the Materials Day Poster Session immediately following the Symposium, award winners were Mechanical Engineering graduate student Erin Looney, Media Arts and Sciences graduate student Bianca Datta, and Materials Science and Engineering Postdoctoral Associate Michael Chon.
The Materials Research Laboratory (MRL) serves interdisciplinary groups of faculty, staff and students, supported by industry, foundations and government agencies to carry out fundamental engineering research on materials. Research topics include energy conversion and storage; quantum materials; spintronics; photonics; metals; integrated microsystems; materials sustainability; solid-state ionics; complex oxide electronic properties; biogels; and functional fibers.
Machine learning coupled with mechanical strain and radiation nudging of atoms promise powerful new control over materials.
|MIT Professor Ju Li speaks at the MIT MRL Materials Day Symposium Oct. 9, 2019. Image, Denis Paiste, Materials Research Laboratory.|
Exerting human will on materials atom-by-atom is an engineer’s dream. “Machine learning will help us get there,” Professor Ju Li, Battelle Energy Alliance Professor of Nuclear Science and Engineering, and Professor of Materials Science and Engineering, says. Today, his work to gain insights into improving materials is advantaged by machine learning and automated experimentation, tools that he suggests will take over much of the work of current day engineers. Li spoke at the MIT MRL Materials Day Symposium Oct. 9, 2019.
Li and colleagues first applied machine learning tools to control elastic strain, which is a combination of stretch, shear and pressure that allows a material to deform reversibly, focusing on nanoscale-sized silicon and diamond. These changes can affect the electronic and optical properties of materials, for example, strained silicon technology uses biaxial elastic strain of about 1 percent to boost electronic current flow by more than 50 percent. “With improvement in sample quality, you can stretch silicon nowadays by more than 10 percent, deeply into what I call the ultra-strength ocean,” Li says.
Research on niobium nanowires drawn inside a nickel-titanium shape memory alloy showed when this material is strained under a load, this material can undergo 6% elastic deformation reversibly up to half a million times, Li relates. Synchroton images show this is truly elastic deformation and not a phase change, he says.
“Niobium is a standard superconductor and with strain you can change its superconducting temperature,” Li says. Elastic strain in these nanowires can change the critical temperature by about 20 percent. “From an engineering point of view, if you have to use liquid helium cooling, this gives you a lot of advantage. And you can also change the critical magnetic field significantly, both the upper critical field and lower critical field.”
“With nanotechnology, we suddenly have a big explosion of samples which can sustain a large dynamic range of elastic strain in tension, in shear, in combined loading,” Li notes. But calculating these effects on materials needs to account for six different strain possibilities, making it an ideal problem to address with machine learning. “Because strain has six components that you can change, it’s difficult to visualize,” he says. This work can reveal, for example, the least-energy pathway of changes in the bandgap of a material in response to mechanical actuation.
Using machine learning and a neural network, Li was able to predict a strain pathway to turn silicon’s electronic properties from semiconducting to metallic. “You really need to be able to go far out in strain energy to be able to hit direct bandgap silicon, but if you hit that, you can make lasers and integrated photonics much more miniaturized, because you don’t need phonons to give off photons.” Li says.
The neural network also identified topological transitions of band structure at certain elastic strains, which allows the researchers to label and visualize them. These new techniques have a wide range of applications from topological quantum computing to solar cell technology. Li’s team applied them to diamond and, he says, “This is the first time we are able to get unobstructed visualization of the six-dimensional ideal strain surface of silicon and diamond.”
In earlier work, experimentalists have demonstrated that they could bend and stretch ultrafine needles of diamond without breaking it. “We can make diamond a direct bandgap material with much less energy than you need from experiment, and we can make the bandgap of diamond become that of gallium nitride,” he says.
“In collaboration with Ming Dao and Subra Suresh, we’re looking at this experiment again and we are basically finding that by judiciously choosing the direction of bending, you can make diamond to have exceedingly small bandgaps. So that has yet to be experimentally validated,” Li adds.
Another line of research focuses on guiding the well-collimated electron-beam in a transmission electron microscope to move individual atoms, such as an individual phosphorus atom on graphene. Experiments show the ability to rotate a carbon-phosphorus bond 180 degrees because of the momentum transfer from the relativistic electron to the phosphorus atom, Li says. “By changing the direction of the electron beam, we can also have the carbon-phosphorus bond rotate 90 degrees, and control clockwise versus counter-clockwise rotation. This flips the concept of radiation damage, because instead of damaging the sample, we can create precisely any atomic structure that we want.”
Using an analogy of moving a phosphorus atom to dribbling a soccer ball around a field, Li, explains, “Sometimes we actually kick out the phosphorus atom, and in that case we have to go back and pick out a new ball from the side of the field and then play again. So there is a tradeoff of throughput of atomic manipulation by electron radiation and the risk of losing the soccer ball, and we can play that game by using machine learning.”
Li continues, “The machine can learn and improve the theoretical calculations, so in the future we have this optimized balance between throughput, which we believe eventually can hit one millisecond per atom step, in other words, you can assuredly move 1,000 atomic steps in one second. And with that, we’ll be able to have what I call atomic engineering. This is the ability to exert human will onto single individual atoms, to dictate their precise location, spin, redox state, etc.”
MIT Materials Research Laboratory 2019 Poster Session winners are Mechanical Engineering graduate student Erin Looney, Media Arts and Sciences graduate student Bianca Datta, and Materials Science and Engineering Postdoctoral Associate Michael Chon.
The Poster Session was held immediately after the Materials Day Symposium on Oct. 9, 2019. Winners, who were selected by non-MIT affiliated attendees, each receive a $500 award.
Media Arts and Sciences graduate student
POSTER: “Simulation-based optimization towards fabrication of bio-inspired nanostructures exhibiting structural coloration.”
Datta is using simulation techniques and rapid prototyping to design surfaces that display color like butterfly wings.
Advisor: Christine Ortiz, Morris Cohen Professor of Materials Science and Engineering
POSTER: “High capacity CMOS-compatible thin film batteries on flexible substrates”
Chon is developing all solid-state flexible microbatteries that combine a germanium anode, a ruthenium dioxide cathode and lithium phosphorous oxynitride (LiPON) solid electrolyte. The thin film batteries can be stacked and folded or incorporated directly into integrated circuits.
Advisor: Carl V. Thompson, Stavros Salapatas Professor of Materials Science and Engineering, and Director, Materials Research Laboratory
Erin E. Looney
Mechanical Engineering graduate student
POSTER: “Machine learning-based classification of environmental conditions for PV module testing and design”
Looney simulates solar cell material operation under real world conditions by combining temperature, solar spectra and humidity data to estimate performance with 95 percent accuracy. She showed that a statistical method called a k-means algorithm can produce these results with 1,000 times fewer data inputs.
Advisor: Tonio Buonassisi, Professor of Mechanical Engineering
– Materials Research Laboratory
Updated October 28, 2019
Annual MIT Materials Day Symposium highlights latest innovations on Oct. 9, 2019.
Machine learning tools are both helping to design new materials and devices and to help those devices run at their best.
Optical spectrometers, for example, are devices that record Iight intensity as a function of wavelength and identify chemicals based on their response to light. MIT Associate Professor of Materials Science and Engineering Juejun (JJ) Hu, last year developed a new chip-based spectrometer that employs an algorithm which improves resolution 100 percent compared to the textbook limits, called Rayleigh limits.
“We developed an algorithm that allows us to extract the information with much better signal-to-noise ratio,” Hu explains. “We have validated the algorithm for many different kinds of spectrum.”
Unlike the conventional shape of glass lenses which are often curved, his new optical devices feature an array of specially designed optical antennas that add a phase delay to the incoming light, which enables many different functions. Hu currently is working with UMass researchers to perfect an algorithm that can screen potential designs for these devices. The algorithm can evaluate the workability of irregular shapes that go beyond conventional shapes likes circles and rectangles.
“The algorithm allows us to train it with existing data,” Hu says. “It can recognize the underlying connections between complex geometries and the electromagnetic response.” The algorithm can find hidden relations much faster than conventional full-scale simulation methods. The algorithm can also screen out potential combinations of materials and functions that just won’t work. “If you use conventional methods, you have to waste lots of time to exhaust all the possible design space and then come to this conclusion, but now our algorithm can tell you really quickly,” he says.
Hu will present his research at the MIT Materials Research Laboratory’s annual Materials Day Symposium on Wednesday, Oct. 9, in Kresge Auditorium. The Symposium runs from 8 a.m. to 3:30 p.m. and is immediately followed by a Poster Session in La Sala de Puerto Rico on the second floor of Stratton Student Center. Register here.
Atlantic Richfield Associate Professor of Energy Studies Elsa A. Olivetti will discuss her work on an artificial-intelligence system that scours through scientific papers to deduce materials science “recipes.” Her team is currently working on experimental verification, particularly focused on catalysts materials.
“We are constantly refining and improving our system from improving overall accuracy to expanding to other parts of the paper, such as results, to other kinds of documents, such as patents,” Olivetti says.
AI can also help to improve sustainability. “If we can know better how to make new materials, we might be able to inform how to make them in a lower resource consuming way,” Olivetti says.
Keynote speaker Dr. Brian Storey, Toyota Research Institute’s Director of Accelerated Materials Design & Discovery, will discuss several collaborative projects focusing on research and development of materials for battery and fuel cell electric vehicles.
Other Materials Day speakers are: Professor Carl V. Thompson, Director, Materials Research Laboratory; Professor Klavs F. Jensen, Departments of Chemical Engineering and Materials Science & Engineering; Professor Asu Ozdaglar, Department Head, Electrical Engineering & Computer Science; Professor Ju Li, Departments of Nuclear Science & Engineering and Materials Science & Engineering; and Assistant Professor Rafael Gomez-Bombarelli, Department of Materials Science & Engineering.
MIT graduate students and postdocs will give two-minute talks on their research during a “Poster Previews” session before the lunch break. The Poster Session runs 3:35 to 5:45 p.m. with an awards presentation at 5:30 p.m.
Transformative new tools to probe atomic structures in action are yielding better designs for metals, solar cells and polymers.
Powerful new combinations of X-rays, electrical probes and analytical computing are yielding insights into problems as diverse as fatigue in steel and stability in solar cells.
“Fatigue in steel is a major issue; you don’t see any changes in the shape of your material, and suddenly it fails," Assistant Professor C. Cem Taşan said during the MIT MRL Materials Day Symposium on Wednesday, Oct. 10, 2018. “We are putting a lot of effort in maintenance and safety, yet still we have devastating accidents,” he said, recalling the airline incident in April 2018 when a jet engine turbine blade broke apart and shrapnel from the engine broke a plane window fatally injuring a passenger.
“The airline company basically said that component passed all the maintenance requirements. So it was checked, and they couldn’t see any kind of fatigue cracks in it,” Taşan, the Thomas B. King Career Development Professor of Metallurgy, explained. Taşan is developing new steel and other metal alloys that are safer, stronger and lighter than those currently available.
Failure in metals is a complex mix of cracks and other changes in the microstructure caused by temperature, bending, stretching, compression and other forces, but most can survive at most one of these impacts before unleashing a cascade of subtle changes that ultimately result in failure.
Design for repair
Taşan outlined progress on a vanadium-based alloy that changes back to its original state when stress is taken away, and a new type of steel that can be transformed back to its original state when heat is applied. Stress tests to measure fatigue in Taşan’s new steel showed improvement over other steels.
Underlying these findings are new nanoscale experimental techniques that Taşan employs to identify the multiple causes of failure in metal alloys. Taşan combines energy-dispersive X-ray spectroscopy and scanning electron and transmission electron microscopes to capture data on tension, bending, compression or nanoindentation of materials. These type of microscopic measurements are called in situ techniques.
Another technique studies how a metal alloy absorbs hydrogen and its effect on the metal. For example, Taşan played movies that show how plastic strain is accommodated to two phases in a high-entropy alloy.
“These techniques allow us to see how the failure process is taking place, and we use these techniques to understand the mechanism of these failure modes and potentially repair mechanisms. Finally, we use this understanding to design new alloys that utilize these mechanisms,” Taşan said. “You are trying to design a mechanism that can be used by the material over and over and over again to deal with the same type of crack that it is facing.”
Taşan’s investigations revealed three different types of crack closure mechanisms in steel: plasticity, phase transformation and crack-surface roughness. “If I want to activate all of these crack closure mechanisms, what I need to do is design a microstructure that is metastable, nano-laminate(d) and multi-phase at same time,” he said. He said the new steel alloy successfully combines all three characteristics.
Materials Research Laboratory Director Carl V. Thompson noted that how a material is made determines its structure and its properties. These properties include mechanical, electrical, optical, magnetic and many other properties. Materials science and engineering encompasses an entire cycle from designing methods for making materials through analyzing their structure and properties, to evaluating how they perform. “Ultimately most people go through this process to make materials that perform in either a new way or in a better way for systems like automobiles, your cell phone, or medical equipment,” Thompson said.
Engineering perovskite solar cells
Silvija Gradečak, Professor in Materials Science and Engineering, addressed the promise and the problems of perovskite solar cells. Hybrid organic-inorganic perovskites, such as methyl ammonium lead iodide, are a class of materials that are named after their crystal structure. “They are potentially lightweight, flexible and inexpensive as photovoltaic devices,” Gradečak said.
However, perovskite solar devices tend to be unstable in water, oxygen exposure, UV irradiation, and under voltage biasing. As many of these changes are dynamic and happen at nanoscale, understanding the structure of these materials can be complemented with information from electrical currents. “By using the electron beam, we can mimic the condition of the electron current within the device,” she said.
Gradečak uses a technique called cathodoluminescence to probe these perovskite materials. “Our cathodoluminescence setup is unique because it enables so-called hyperspectral imaging. It means that the full optical signal is detected in each point of the complementary structural image. As the beam interacts with the sample, we are detecting light, and we do this as the electron beam moves across the sample. That is specifically important for samples that are unstable as they are irradiated with the electron beam,” she says.
This technique revealed that perovskite material examined under an electron microscope while applying a voltage to the sample for 1 minute resulted in a dramatic current increase in the material. “That also corresponds to the I/V (current/voltage) measurements outside of the scanning electron microscope that we performed,” she said. When the voltage bias is removed, the sample relaxes back to its initial state.
“What we think is really happening is that by biasing, there are ions that are moving and they agglomerate at the edges of the sample or at the grain boundaries, and after you remove the bias, they will relax back,” Gradečak said.
Work in Gradečak’s group by Olivia Hentz (PhD ’18) combined photoluminescence data with Monte Carlo simulations to extract mobility of the defects that are moving. “More interesting, and how we can apply this method, is to understand how the material’s properties are influenced by synthesis. If you synthesize the material and you change, for example, the grain size, we can think about whether these ions that are moving will have different mobilities inside of the grain versus along the grain boundaries,” Gradečak said.
Hentz found that the mobility at the grain boundaries is 1,500 times faster than in the bulk. “The ions do move in the material, they move under the biasing conditions and that mobility is very different inside of the grain and along the grain boundaries,” Gradečak said. “By engineering the material and engineering the grain size, one can influence by how much the material will be influenced during the device operation. And this result correlates with the fact that single crystalline perovskite materials are significantly more stable than polycrystalline ones.”
Transformative new tools
In the Keynote address, BP Amoco Chemical Company Senior Research Chemist Dr. Matthew Kulzick detailed new X-ray technologies and sample chambers that are yielding insights into fighting metal corrosion, improving catalytic reactions and more. “The current evolution of tools is spectacular,” he said, noting the stunning images at 20-nanometer scale showing highly localized composition of materials.
MIT Nuclear Reactor Lab Director David E. Moncton discussed advances in X-ray tubes, noting that current versions of small scale X-ray tubes are about 100 times better than those of 100 years ago. X-ray source brilliance is increasing at two times Moore’s Law, which predicted the exponential growth of transistors in silicon chips, he noted.
Still Synchroton sources such as the Advanced Photon Source a national user facility at Argonne National Laboratory, offer beam brilliance that is 12 orders of magnitude higher than X-ray tubes. “Advanced X-ray capability is the most important missing probe of matter at nano centers and materials research labs that are not located at synchrotron facilities,” he said.
Compact X-ray free-electron laser devices hold the promise of bringing synchrotron-like examination capabilities to campus research labs, Moncton said. Moncton, who was the founding director of the Advanced Photon Source, is collaborating with Associate Professor William S. Graves at Arizona State, which is home to world’s first compact X-ray free-electron laser (CXFEL).
“The emittance is very similar to a synchrotron source,” Moncton said. “If you built a compact X-ray FEL on this compact source platform, it would outperform today’s synchrotron facilities by a number of orders of magnitude.”
X-ray phase contrast imaging has also advanced microscopy, Moncton said, displaying an image showing air bubbles in the lungs of a fruit fly. Pump-probe techniques enable studies of biological proteins performing bio-chemical processes in real time.
“Having a local synchrotron-like source would be revolutionary,” Moncton said.
Less damaging microscope
Professor of Electrical Engineering Karl Berggren described his efforts to develop a new type of electron microscope based on the quantum character of electrons to improve microscopy. One of the goals is to reduce radiation damage to biological samples from imaging them.
With support from the Gordon and Betty Moore Foundation, Berggren is collaborating on this research with Professor of Physics Mark Kasevich at Stanford University in California, Professor of Physics Peter Hommelhoff at the Friedrich Alexander University, Erlangen-Nürnberg, in Germany, and Professor of Physics Pieter Kruit at the Technical University of Delft in the Netherlands. “What we’d like to do is basically try to take advantage of the counter-intuitive quantum properties of electrons,” Berggren said.
In one approach, he employs a series of electron beam splitters and mirrors to improve the performance of scanning electron microscopes. “What we’re doing now is essentially making a test bed by which we can develop all the electron optics to try to put together a machine,” Berggren said. Along the way, his group has developed a microscope that lets you image the top and bottom of a sample at the same time.
“We know that electrons at high voltage will pass through many samples with interacting with just a small phase shift,” he said. “In fact, we want to work in that limit for imaging bio molecules.” The right combination of beam splitters could reduce electron-induced damage to the sample by 100 times, he said.
Dr. Frances M. Ross, formerly of the Research Division at the IBM T. J. Watson Research Center and a new arrival at the Department of Materials Science and Engineering this academic year, described her observations of nanowire growth in an electron microscope. This vapor-liquid-solid process was first described in 1964, but the atomic-level details of how the nanowires grow could not be observed until recent improvements in electron microscopy technique.
|Movie shows the growth of a silicon nanowire (lower region) from a catalytic droplet of gold silicon (AuSi) liquid (dark hemisphere above). Growth takes place by rapid addition of planes of silicon atoms at the catalyst/silicon interface. The nanowire diameter is 50 nanometers and growth took place at 500oC. Video courtesy of Frances M. Ross. Reproduced from Chou et al., “Nanowire growth kinetics in aberration corrected environmental transmission electron microscopy,” Chem. Commun., 2016, 52, 5686-5689, with permission from The Royal Society of Chemistry."|
Showing a movie of a silicon nanowire growing from a gold-silicon catalyst droplet, Ross said, “To grow these silicon nanowires, we just put gold on silicon and heat it up. The gold and silicon automatically form droplets, in the same way that water forms droplets on a sheet of glass.” When additional silicon is then supplied, the droplets act as a catalyst and a silicon nanowire grows from each droplet. “Nanowire growth illustrates the fact that we can get a self-assembly process that is intrinsically very simple to form a structure that can be quite complex,” Ross explained. “You can see features like the atomic level structure of the nanowire and catalyst, the effect of temperature and gas environment, and even the dynamics of the growth interface and how the catalyst really works.” The silicon nanowire grows in little jumps despite a steady flow of source material, she noted, providing detailed information on the pathways by which the atoms assemble into the nanowire.
Adding nickel to this process resulted in a nickel disilicide particle embedded in the silicon nanowire – a quantum dot. “You almost expect to see unexpected things because the movies capture every point along the way as the material evolves,” Ross said. “In situ microscopy is really the only way to get these type of detailed relations between the structure, the properties and even the catalytic activity of individual nanoscale objects.”
“We’re in a very exciting time for electron microscopy, where advances in instrumentation are helping us understand materials growth at the atomic scale,” Ross said.
Uncovering crystal structure
James LeBeau, Visiting Professor of Materials Science and Engineering, explained that scanning transmission electron microscopy provides direct imaging of atomic structure using an extremely small (< 1x10-10 m) electron probe. LeBeau uses the scanning transmission electron microscope to develop and apply new ways to characterize atomic structure of materials to understand their properties. Further, he is applying machine learning to control the microscope, using an approach similar to that used to enable self-driving cars to recognize signs and lane lines.
Beyond imaging, “we can also acquire a full chemical spectrum at every single point in our dataset. This allows us to not only directly determine which atoms are in the material, but their bonding configuration as well,” LeBeau explained. He displayed an image showing lanthanum atoms sharing a sub-lattice with strontium and aluminum sharing a sub-lattice with tantalum. “These datasets become directly interpretable. You see the chemistry,” he said.
“We can even use this data to measure the atomic scale electric field,” LeBeau said, showing an image in which the color represents the electrostatic field vector and the intensity of the color represents its magnitude. LeBeau also was able to use these techniques to uncover the particular crystal structure of ferroelectric hafnium dioxide (HfO2). The atomic scale insights are critical as hafnium dioxide is compatible with silicon processing technology, which will pave the way for new memory applications. “By combining different types of data, we can explain the origin or ferroelectricity in these films and really rule out alternative explanations,” he said.
Twenty graduate students and postdocs gave two-minute previews during the Materials Day Symposium, which was immediately followed by a Poster Session. In all, 60 presented research posters in La Sala de Puerto. The winning presenters were graduate students Vera Schroeder, Rachel C. Kurchin, Gerald J. Wang and Philipp Simons, and Postdoctoral Associate Mikhail Y. Shalaginov.
BP chemist details new X-ray technologies and sample chambers that are yielding insights into fighting metal corrosion, improving catalytic reactions and more.
|BP Amoco Chemical Company Senior Research Chemist Dr. Matthew Kulzick outlines advances in imaging technology during the MIT MRL Materials Day Symposium on Wednesday, Oct. 10, 2018. Photo, Denis Paiste, Materials Research Laboratory.|
New electron microscopy techniques can help solve corrosion problems that are worth millions of dollars to industrial companies, BP Amoco Chemical Company Senior Research Chemist Dr. Matthew Kulzick told the MIT MRL Materials Day Symposium on Wednesday, Oct. 10, 2018.
“Materials Science is critical. It’s really material in the financial sense,” Kulzick said. “Solutions demand timely and accurate information. If I’m going to solve a problem, I’ve got to know what’s actually going on, and to do that I need all of these different interrelated tools to be able to go in and find out what’s happening in systems that are important to us.”
New X-ray technologies and sample chambers are producing stunning images at 20-nanometer scale showing highly localized composition of materials. “The current evolution of tools is spectacular,” he said.
Beginning in 2003, Kulzick built a new inorganic characterization capability for BP Amoco Chemical Company, MRL Associate Director Mark Beals said in introducing Kulzick. Kulzick has been working with Nestor J. Zaluzec, a senior scientist at Argonne National Laboratory, as well as with the BP International Center for Advanced Materials [ICAM], whose partners include the University of Manchester, Imperial College London, the University of Cambridge, and the University of Illinois Urbana–Champaign.
He outlined advances in imaging technology such as the π Steradian Transmission X-ray Detection System developed at the U.S. Department of Energy’s Argonne National Laboratory and advances in sample holder technology that BP developed collaboratively with Protochips that allow analysis of materials in gas or liquid filled chambers. Microscopic measurements using these holders, or cells, which can include micro-electro-mechanical systems (MEMS), are called in situ techniques.
“A number of years ago we worked with Protochips, and we modified that holder technology to allow the X-rays coming out of that system to get to the detector,” Kulzick explained. Images of a palladium and copper-based automotive catalyst from four different generations of Energy-dispersive X-ray technology illustrated the evolution from images lacking in detail to a nanoscale compositional image acquired in just 2.5 seconds that shows the location of palladium in the chemical structure. “So it’s really transformative in understanding what’s happening chemically at the nanoscale,” Kulzick says.
Placing a closed cell filled with hydrogen gas to simulate reduction of the catalyst inside a transmission electron microscope produced images that showed palladium particles remained unaffected while copper particles either migrated toward palladium particles or clustered together with other copper particles. “We can actually observe the changes that are happening in that localized area under reduction, and this is extremely important if we really want to understand what’s happening,” Kulcizk says. “All of that diversity is occurring in what amounts to roughly a square micron of area on the surface of the material.”
|BP Amoco Chemical Company Senior Research Chemist Dr. Matthew Kulzick addresses the MIT MRL Materials Day Symposium on Wednesday, Oct. 10, 2018. Photo, Denis Paiste, Materials Research Laboratory.|
“Just imagine what I could do with this kind of technology with regard to understanding how to activate a catalyst, how to regenerate a catalyst,” Kulzick said.
Techniques developed by Prof. M. Grace Burke at the University of Manchester in the UK allow observation of chemical changes in a piece of metal over a period of hours such as dissolving a manganese sulfide inclusion from a small piece of stainless steel soaking in water, Kulzick said. “This proved a point for her with regards to corrosion mechanisms that are relevant in the nuclear industry where they worry about what’s initiating crack formation and which she has argued for years that attack of the manganese sulfide by water was one of the underlying mechanisms,” he said.
Direct electron capture cameras
A significant advance for analyzing organic materials is direct electron capture cameras, Kulzick said. “One of the problems with bombarding things with electrons is beam damage, so you want to use as little as you can with the right energies. The direct electron capture cameras allowed us to reduce that dose,” he said.
For example, Qian Chen, Assistant Professor of Materials Science and Engineering, at the University of Illinois Urbana–Champaign, has used this enhanced sensitivity and lower dose radiation to a do a series of images at differing tilts to generate a three-dimensional image of a polymer membrane. Computational image analysis becomes important with these 3D structural images. “Without the ability to digitize that material like we’ve done, we would never be able to understand this diversity of structure and make it more rational,” Kulzick said.
Further analysis of the polymer membrane – soaked in a solution of zinc and lead – with Analytical Electron Microscope (AEM) techniques developed by Zaluzec at Argonne National Laboratory revealed that different ions enter into the polymer membrane at different locations. The next step is to understand how ions interact with the membrane structure and how that impacts permeation in the systems, Kulzick said. Chen also analyzed the polymer membrane in water inside a graphene cell, he said, and that work showed swelling of the membrane.
“We hope to put all these pieces together and form a really detailed understanding of how a system like this functions,” he said.
|MIT Materials Day Poster Session winners are [left to right] graduate students Vera Schroeder, Rachel C. Kurchin, Gerald J. Wang and Philipp Simons, and Postdoctoral Associate Mikhail Y. Shalaginov. Sixty students and postdocs presented their posters in La Sala de Puerto on Wednesday, Oct. 10, 2018. Of those, 20 gave two-minute poster previews during the Materials Day Symposium immediately before the Poster Session. Photo, Denis Paiste, MIT Materials Research Laboratory.|
The external advisory board dinner will be held on October 9, 2019. Immediately following the Materials Day Poster Session.
Location: MIT Student Center, West Lounge
Cocktails will start at 6:30pm.
The advisory board meeting will be held on October 10, 2019.
Location: Bush Room, Building 10-105
8:30am - 4:30pm
Materials Day is scheduled for October 9, 2019
Poster Setup will be in the Student Center - La Sala de Puerto Rico
REGISTRATION IS NOW CLOSED, however if you'd still like to present a poster, please feel free to show up with your poster and you will be assigned a poster board.
Please include the MRL logo on the top left side of your poster. Download the MRL logo.
Posters may be set up between 12:00 pm and 3:00 pm the day of the event. Individuals are expected to be with and remain with their poster during the Poster Session, from 4:00-6:00pm.