The MEMS device was fabricated from a single chip of 200-μm-thick silicon. The reverse side of the wafer was first coated with 2.5 μm of plasma-enhanced chemical vapour deposition (PECVD) SiO . A 100-nm coating of chromium was next deposited on the top surface of the silicon using a thermal evaporator. The MEMS device pattern was created in a layer of positive photoresist using a g-line photolithography process. The mask was a ‘halo’ design31 that is, instead of etching away all of the unwanted areas of silicon, trenches were used in an outline of the structure, to keep a constant etch rate and profile over all etched areas. The halo was 10 μm wide. The photoresist pattern was then used as a mask to wet-etch the chrome using a nitric acid chrome etchant for 100 s, thus etching the MEMS device proof mass pattern into the chrome. The resist was then removed ultrasonically with acetone and isopropanol, leaving the chrome etch mask in place. A 7-μm layer of AZ-4562 photoresist was then spun onto the back of the sample and used later to make the sample free-standing. The sample was fixed to a carrier wafer (chrome side up) using a thin, spun-on layer of Crystalbond 509 (as mounting adhesive) in solution with acetone. To ensure a good thermal contact the sample was weighted and left on the hotplate at 88° C (just above the melting point of Crystalbond 509) for 5 min. The sample was next placed in an Oxford Instruments PlasmaPro 100 Estrelas Deep Silicon Etch System, and Bosch-etched32 for 80 min using an SF , C F process optimized for highly anisotropic trenches. This etch was the same depth as the silicon and stopped when it reached the SiO back layer. The PlasmaPro 100 Estrelas Deep Silicon Etch System allows control of the gas flow, enabling processes to be tuned with negative and positive defined etch profiles. Our spring profiles are vertical to within 0.5°. To remove the sample from the carrier wafer it was heated to 88° C for 5 min, and then pushed laterally off the Crystalbond 509, which is now fluid. The SiO and the AZ–4562 layers enabled this to be done without damaging the MEMS device structure. The sample was then turned upside down and placed (not affixed) on a blank piece of silicon. The residual Crystalbond 509 and photoresist were removed from the bottom of the sample using an O plasma ash. The sample was exposed to a CF /O etchant plasma until all of the SiO was removed, making the sample free-standing. Our MEMS device is comprised of a proof mass suspended from three curved cantilevers/flexures. To better understand the physical characteristics of this system we first discuss these flexures individually. Consider a cantilever, clamped at one end, and free to move at the other. A proof mass mounted on the moving end will oscillate with a frequency that depends on the geometry of the cantilever, and the Young’s modulus of the material from which it is made. The proof mass will oscillate along an arc defined by the length of the flexure. The system will behave as a Hooke’s law spring, with a linear relationship between force and displacement. This behaviour can be observed in Extended Data Fig. 4a. A curved single cantilever also behaves in the same manner, as seen in Extended Data Fig. 4b. To create an anti-spring, one can take two such curved cantilevers and attach them at a central pivot point. A proof mass mounted at this point will no longer be able to trace out an arc as it oscillates. Instead, because of the symmetrical forces applied by the two identical cantilevers, its motion will be constrained along a vertical axis (as presented in Fig. 1). It is this constraint that causes the spring constant to change as the displacement increases. Instead of observing a linear relationship between force and displacement, a nonlinear behaviour is observed (see Extended Data Fig. 4c). This now means that the spring gets softer with increasing displacement. A four-flexure anti-spring system is a simple extension of a two-flexure system. Here, a second pair of cantilevers are placed below the first pair, which allows a non-point-source proof mass to be suspended. The behaviour of the spring is still nonlinear, and is displayed in Extended Data Fig. 4d. The behaviour is identical to that of a two-flexure system, except the system can support twice the mass. Both the two- and four-flexure anti-spring systems can be used to create oscillators that have low resonant frequencies. When the limits of k/m are pushed to create the lowest resonant frequency possible, however, these systems become unstable. They become unstable because the motion is so well constrained along its vertical axis that the spring gets softer and softer until it can no longer support the weight of the proof mass. This behaviour can be observed in Extended Data Fig. 4c and d: as the force increases, the displacement increases rapidly. A stable resonant frequency is imperative for a useful relative gravimeter, so this instability would create problems if used for the design of a MEMS gravimeter. It would require the use of a closed-loop feedback system. Our MEMS device utilizes a novel three-flexure anti-spring system, with one flexure of the upper pair of cantilevers removed (see Fig. 1). In the first instance, the device behaves as a four-flexure anti-spring: it gets rapidly softer as the displacement of the proof mass increases. The anti-spring behaviour is maintained while the proof mass moves along its vertically constrained axis. The asymmetry of the system, however, means that the device does not stay constrained along the anti-spring constraining axis. The single upper flexure ultimately tilts the proof mass marginally away from the constraining axis. As the motion is pulled from this axis, the anti-spring trend is halted and the device regains a Hooke’s law behaviour, where dF/dz is a constant. This behaviour can be observed in Extended Data Fig. 4e, where the gradient of force versus displacement reaches a minimum at z = 0.6. This means that the device assumes a constant value of k at the minimum stiffness value that we have demonstrated to be stable over many months (as demonstrated by Extended Data Fig. 9). The proof mass motion is measured using an optical shadow sensor14. Using a fused silica ‘C’-shaped support structure, a red LED (powered at 0.3 mW) was shone onto a split photodiode, with the MEMS device proof mass mounted in between. The change in intensity incident on the photodiode resulting from the motion of the proof mass shadow was then used as a measure of the motion. The split photodiode was made from two 5 mm by 10 mm planar silicon photodiodes, and wired to give a differential output. A split photodiode was used so that at the nominal position of the proof mass the output signal was zero. This allowed maximal amplification without saturation of the measurement instrumentation. The LED signal was modulated (at a frequency of 107 Hz with a 50:50 duty cycle) to reduce the 1/f noise in the output signal. The modulation was carried out by turning the LED on and off with an HP 33120A square-wave signal generator. A precision current-stabilizing resistor (displayed in Fig. 2) maintained the LED drive current; this resistor was heat sunk to the fused silica ‘C’-shaped structure. The current output from the photodiode was first converted into a voltage using a Stanford Research Systems SR570 current-to-voltage converter, band-passed between 3 Hz to 100 Hz, and amplified by a factor of 106 V A−1. This amplified signal was then de-modulated via an analogue lock-in amplifier (Femto LIA-MV-200) referenced from the signal generator. The lock-in amplified the signal with a gain of ten and undertook readings with a time constant of 3 s. This analogue signal was passed through a Stanford Research Systems SR560 low pass filter of 0.03 Hz, 12 dB per octave, to remove aliasing and filter seismic noise, before being digitized via a 16-bit, analogue-to-digital converter (National Instruments M Series 6229) and recorded by a computer with a 24-s time constant. Analogue signals were used to reduce digitization noise that would have occurred if a digital signal had been amplified by this magnitude. The shadow sensor has a read-out noise floor of ≤10 μGal at the sampling frequency of 0.03 Hz, and a dynamic range of about 50 μm. A large dynamic range is required because of the large initial displacement (0.8 mm) of the proof mass when it is tilted to its vertical operating orientation, thus making initial alignment of the MEMS device difficult. Although the maximum peak-to-peak displacement of the proof mass caused by the tides is only 16 nm, the proof mass also oscillates at its resonant frequency by up to 100 nm owing to seismic ground motion. A high dynamic range is also useful to measure this signal, which is ultimately removed from the data by averaging with a 0.03 Hz filter in the read-out electronics. The control loops used to maintain the temperature of the system were proportional integral derivative control mechanisms, written in Labview (http://www.ni.com/labview/). Temperatures were monitored using a four-terminal measurement of small platinum resistors, via two Keithley 2000 digital multimeters. A four-terminal measurement eradicates contact resistance by driving the thermometer with a current and measuring the voltage across it. This removes the temperature sensitivity of external wires. Low-temperature-coefficient Manganin alloy wires were used for these connections to minimize parasitic thermal conduction. One platinum resistor was placed on the outer frame of the MEMS device and three were placed equidistantly around the copper shield. Wire wound resistors were used as the heating mechanism to feedback into the system; again, one of these was placed on the MEMS device frame and three around the shield. The output signal to the heaters was sent via a National Instruments (USB 6211) digital-to-analogue converter (DAC) card, and the heaters were powered with non-inverting amplifiers with a capability to power up to 100 mA. All circuitry and instrumentation used to amplify and measure the output signal, and to measure and control the system temperature, were selected for their high thermal stability. This entire configuration was constructed in a vacuum chamber with a pressure of ≤10−5 mTorr. Although proportional integral derivative (PID) temperature control was implemented for the MEMS device and the shield, there were other components with variations that could not be actively controlled. These were the room temperature that coupled into the data via a temperature-sensitive lock-in amplifier, and the intensity variations of the LED, which were monitored using a monitor photodiode. There was also an offset, and a linear drift of under 150 μGal per day once the system had been left evacuated for over a week. This drift term is due to stress in the silicon flexures. Like all mechanical systems, application of stress leads to anelasticity, which causes creep and drift over long timescales. Our device also shows polynomial drift which decays away approximately one week after evacuating the apparatus. The polynomial drift is probably due to adsorbed water on the surface layer of silicon, and could be mitigated by baking out the system before evacuation. Extended Data Fig. 5 demonstrates this initial polynomial drift. The data were therefore regressed against the temperature measurements listed above, the drift offset and the intensity. This regression—carried out in Matlab (http://uk.mathworks.com/products/matlab/) with the mregg tool—identified correlations between the output data and these parameters, and removed any resulting correlated trends from the final data. Floor tilt and power variation of the LED were also monitored, but neither had any discernible effect on the signal and were therefore not regressed. The correlation coefficient, R, between the averaged theoretical and experimental tide data was calculated using Matlab’s corrcoeff function. An R value of 0.86 was produced for the plot presented in Fig. 3. To check the level of statistical significance of our experimental data we compared it to the correlation of the noise alone. We created 10,000 random permutations of our data set and calculated the correlation coefficient for each with respect to the theoretical data. This set of R values was plotted as a histogram. This histogram had a distribution with a mean value of zero and a standard deviation of 0.008. The R value from the un-randomized data are 114σ from this distribution, suggesting the correlation is real to an extremely high degree of confidence. Extended Data Fig. 6 is a plot of the root-mean-square acceleration sensitivity of the device over its full spectral range. The tide signal can be observed at 1 × 10−5 Hz. The peak at 10−3 Hz is an artefact of the temperature servo. Between 0.1 Hz and 0.2 Hz the microseismic peak can be recognized; its presence indicates that the device is also a sensitive seismometer. Past observations—made in Scotland from February to March 2000—of the microseismic peak28 confirm the validity of our observation. At 2.3 Hz the primary resonant mode of the MEMS device generates a large peak due to excitation from seismic noise. This plot was used to calculate the sensitivity of the MEMS device. To find a sensitivity in microgal per hertz1/2, it is only necessary to read off the acceleration sensitivity at the point where the data crosses 1 Hz on the horizontal axis. We believe that the value of 40 μGal Hz−1/2 is an overestimate of the true sensitivity of the device because at 1 Hz the influence of both the primary resonance of the device and the micro-seismic peak are important. Although tilt did not have an effect on the tide measurement, we are interested to know at what point tilt would become an issue. Extended Data Fig. 8 presents two plots of an experiment used to assess the effect of tilt on our device. Inside the vacuum tank, the MEMS device was mounted vertically and aligned with the tilt sensor. The y axis of the tilt sensor was aligned with the plane of the MEMS device, with the x axis perpendicular to this (see Fig. 1). Extended Data Fig. 8a demonstrates the induced tilt of the tank and the output of the MEMS device along the x axis. Extended Data Fig. 8b shows the same data as in Extended Data Fig. 8a, but for the y axis. There is a strong correlation between the y-axis variation and the voltage output, giving a tilt sensitivity in this axis of 21.2 μGal per arcsecond. There is less sensitivity to the x-axis tilt with a tilt sensitivity of only 0.6 μGal per arcsecond. The x-axis tilt sensitivity is low because in the vertical configuration the spring resumes a Hooke’s law response, as observed in Extended Data Fig. 1, for which the x-axis tilt variation is plotted against the resonant frequency (the acceleration sensitivity of the device is proportional to the square of the resonant frequency). Ultimately the spring could be tuned to operate with even less variation with tilt in this axis if it were positioned to operate at one of its minima. Alternatively the flexures could be made marginally thicker to shift the minimum in resonant frequency to 90°; this was not carried out because the device did not show sufficient tilt sensitivity to cause concern. The y-axis variation is larger because the device has a mode of oscillation in which the proof mass tilts and pivots about the upper cantilever flexure. When vertical, the device would need to be levelled with an accuracy limited by the y-axis sensitivity (that is, less than 2 arcsec to maintain the current sensitivity) to make repeatable measurements in different locations. This accuracy of levelling is achievable with a simple surveyor’s bubble level. Extended Data Fig. 9 demonstrates two short data sets separated by nearly four months. These were used as a test of the temporal stability of the device. To convert the raw voltage output of the device into a unit of acceleration, a calibration factor was required. By comparing the experimental (blue line) data in Extended Data Fig. 9a with that in Extended Data Fig. 9b we were able to test whether the calibration factor had drifted over time. The same calibration factor has been used to make both of these plots. By averaging the data and changing the calibration factor of Extended Data Fig. 9b, it was found that a change in the calibration factor of 5% made the fit to the tide theory (red line) data noticeably worse. Changes smaller than this were not resolvable. We therefore believe that if the calibration factor has changed, it has done so by no more than 5%. During this period, the vacuum tank was vented and evacuated several times, and the MEMS was moved around each time. This is an important feature of a device that could eventually be used in the field. MEMS gravimeters have many industrial applications. Given their small size and low cost, they could be used for down-borehole exploration in the oil and gas industry33 and used to monitor well drainage. Such devices could also be used for environmental monitoring, where networks of sensor arrays could monitor subsurface water levels34, or to determine the location of historic landfill sites. In the security industry, low-cost and small-size gravimeters would also be useful in detecting subterranean tunnels35, 36 or for imaging of cargo containers, where high spatial resolution via numerous sensors is an advantage37. MEMS gravimeters could also be used in civil engineering. For example, at present in the UK, for many cities built in the Victorian period the placement of utilities is accurate on maps only to within 15 m of landmarks such as trees, fences or buildings. There have been trials of the Scintrex CG5 and MEMS-based arrays should improve mapping resolution. Gravimetry is already used in volcanology and could help to predict eruptions using networks of small, low-cost gravimeter arrays22, 24, 25. A field prototype is currently being developed in Glasgow that will be the size of a tennis ball and require a power supply of under 1 W. A powerless getter pump will be used to maintain vacuum, both the thermal control and the optical read-out will be on-chip; tilt levelling will be included, and all of the read-out and control software will be run on a micro-controller.
The latest buzz in the information technology industry regards “the Internet of things” — the idea that vehicles, appliances, civil-engineering structures, manufacturing equipment, and even livestock would have their own embedded sensors that report information directly to networked servers, aiding with maintenance and the coordination of tasks. Realizing that vision, however, will require extremely low-power sensors that can run for months without battery changes — or, even better, that can extract energy from the environment to recharge. Last week, at the Symposia on VLSI Technology and Circuits, MIT researchers presented a new power converter chip that can harvest more than 80 percent of the energy trickling into it, even at the extremely low power levels characteristic of tiny solar cells. Previous ultralow-power converters that used the same approach had efficiencies of only 40 or 50 percent. Moreover, the researchers’ chip achieves those efficiency improvements while assuming additional responsibilities. Where most of its ultralow-power predecessors could use a solar cell to either charge a battery or directly power a device, this new chip can do both, and it can power the device directly from the battery. All of those operations also share a single inductor — the chip’s main electrical component — which saves on circuit board space but increases the circuit complexity even further. Nonetheless, the chip’s power consumption remains low. “We still want to have battery-charging capability, and we still want to provide a regulated output voltage,” says Dina Reda El-Damak, an MIT graduate student in electrical engineering and computer science and first author on the new paper. “We need to regulate the input to extract the maximum power, and we really want to do all these tasks with inductor sharing and see which operational mode is the best. And we want to do it without compromising the performance, at very limited input power levels — 10 nanowatts to 1 microwatt — for the Internet of things.” The prototype chip was manufactured through the Taiwan Semiconductor Manufacturing Company's University Shuttle Program. The circuit’s chief function is to regulate the voltages between the solar cell, the battery, and the device the cell is powering. If the battery operates for too long at a voltage that’s either too high or too low, for instance, its chemical reactants break down, and it loses the ability to hold a charge. To control the current flow across their chip, El-Damak and her advisor, Anantha Chandrakasan, the Joseph F. and Nancy P. Keithley Professor in Electrical Engineering, use an inductor, which is a wire wound into a coil. When a current passes through an inductor, it generates a magnetic field, which in turn resists any change in the current. Throwing switches in the inductor’s path causes it to alternately charge and discharge, so that the current flowing through it continuously ramps up and then drops back down to zero. Keeping a lid on the current improves the circuit’s efficiency, since the rate at which it dissipates energy as heat is proportional to the square of the current. Once the current drops to zero, however, the switches in the inductor’s path need to be thrown immediately; otherwise, current could begin to flow through the circuit in the wrong direction, which would drastically diminish its efficiency. The complication is that the rate at which the current rises and falls depends on the voltage generated by the solar cell, which is highly variable. So the timing of the switch throws has to vary, too. To control the switches’ timing, El-Damak and Chandrakasan use an electrical component called a capacitor, which can store electrical charge. The higher the current, the more rapidly the capacitor fills. When it’s full, the circuit stops charging the inductor. The rate at which the current drops off, however, depends on the output voltage, whose regulation is the very purpose of the chip. Since that voltage is fixed, the variation in timing has to come from variation in capacitance. El-Damak and Chandrakasan thus equip their chip with a bank of capacitors of different sizes. As the current drops, it charges a subset of those capacitors, whose selection is determined by the solar cell’s voltage. Once again, when the capacitor fills, the switches in the inductor’s path are flipped. “In this technology space, there’s usually a trend to lower efficiency as the power gets lower, because there’s a fixed amount of energy that’s consumed by doing the work,” says Brett Miwa, who leads a power conversion development project as a fellow at the chip manufacturer Maxim Integrated. “If you’re only coming in with a small amount, it’s hard to get most of it out, because you lose more as a percentage. [El-Damak’s] design is unusually efficient for how low a power level she’s at.” “One of the things that’s most notable about it is that it’s really a fairly complete system,” he adds. “It’s really kind of a full system-on-a chip for power management. And that makes it a little more complicated, a little bit larger, and a little bit more comprehensive than some of the other designs that might be reported in the literature. So for her to still achieve these high-performance specs in a much more sophisticated system is also noteworthy.”
An exotic material called gallium nitride (GaN) is poised to become the next semiconductor for power electronics, enabling much higher efficiency than silicon. In 2013, the Department of Energy (DOE) dedicated approximately half of a $140 million research institute for power electronics to GaN research, citing its potential to reduce worldwide energy consumption. Now MIT spinout Cambridge Electronics Inc. (CEI) has announced a line of GaN transistors and power electronic circuits that promise to cut energy usage in data centers, electric cars, and consumer devices by 10 to 20 percent worldwide by 2025. Power electronics is a ubiquitous technology used to convert electricity to higher or lower voltages and different currents — such as in a laptop’s power adapter, or in electric substations that convert voltages and distribute electricity to consumers. Many of these power-electronics systems rely on silicon transistors that switch on and off to regulate voltage but, due to speed and resistance constraints, waste energy as heat. CEI’s GaN transistors have at least one-tenth the resistance of such silicon-based transistors, according to the company. This allows for much higher energy-efficiency, and orders-of-magnitude faster switching frequency — meaning power-electronics systems with these components can be made much smaller. CEI is using its transistors to enable power electronics that will make data centers less energy-intensive, electric cars cheaper and more powerful, and laptop power adapters one- third the size — or even small enough to fit inside the computer itself. “This is a once-in-a-lifetime opportunity to change electronics and to really make an impact on how energy is used in the world,” says CEI co-founder Tomás Palacios, an MIT associate professor of electrical engineering and computer science who co-invented the technology. Other co-founders and co-inventors are Anantha Chandrakasan, the Joseph F. and Nancy P. Keithley Professor in Electrical Engineering, now chair of CEI’s technical advisory board; alumnus Bin Lu SM ’07, PhD ’13, CEI’s vice president for device development; Ling Xia PhD’12, CEI’s director of operations; Mohamed Azize, CEI’s director of epitaxy; and Omair Saadat PhD ’14, CEI’s director of product reliability. While GaN transistors have several benefits over silicon, safety drawbacks and expensive manufacturing methods have largely kept them off the market. But Palacios, Lu, Saadat, and other MIT researchers managed to overcome these issues through design innovations made in the late 2000s. Power transistors are designed to flow high currents when on, and to block high voltages when off. Should the circuit break or fail, the transistors must default to the “off” position to cut the current to avoid short circuits and other issues — an important feature of silicon power transistors. But GaN transistors are typically “normally on” — meaning, by default, they’ll always allow a flow of current, which has historically been difficult to correct. Using resources in MIT’s Microsystems Technology Laboratory, the researchers — supported by Department of Defense and DOE grants — developed GaN transistors that were “normally off” by modifying the structure of the material. To make traditional GaN transistors, scientists grow a thin layer of GaN on top of a substrate. The MIT researchers layered different materials with disparate compositions in their GaN transistors. Finding the precise mix allowed a new kind of GaN transistors that go to the off position by default. “We always talk about GaN as gallium and nitrogen, but you can modify the basic GaN material, add impurities and other elements, to change its properties,” Palacios says. But GaN and other nonsilicon semiconductors are also manufactured in special processes, which are expensive. To drop costs, the MIT researchers — at the Institute and, later, with the company — developed new fabrication technologies, or “process recipes,” Lu says. This involved, among other things, switching out gold metals used in manufacturing GaN devices for metals that were compatible with silicon fabrication, and developing ways to deposit GaN on large wafers used by silicon foundries. “Basically, we are fabricating our advanced GaN transistors and circuits in conventional silicon foundries, at the cost of silicon. The cost is the same, but the performance of the new devices is 100 times better,” Lu says. CEI is currently using its advanced transistors to develop laptop power adaptors that are approximately 1.5 cubic inches in volume — the smallest ever made. Among the other feasible applications for the transistors, Palacios says, is better power electronics for data centers run by Google, Amazon, Facebook, and other companies, to power the cloud. Currently, these data centers eat up about 2 percent of electricity in the United States. But GaN-based power electronics, Palacios says, could save a very significant fraction of that. Another major future application, Palacios adds, will be replacing the silicon-based power electronics in electric cars. These are in the chargers that charge the battery, and the inverters that convert the battery power to drive the electric motors. The silicon transistors used today have a constrained power capability that limits how much power the car can handle. This is one of the main reasons why there are few large electric vehicles. GaN-based power electronics, on the other hand, could boost power output for electric cars, while making them more energy-efficient and lighter — and, therefore, cheaper and capable of driving longer distances. “Electric vehicles are popular, but still a niche product. GaN power electronics will be key to make them mainstream,” Palacios says. In launching CEI, the MIT founders turned to the Institute’s entrepreneurial programs, which contributed to the startup’s progress. “MIT's innovation and entrepreneurial ecosystem has been key to get things moving and to the point where we are now,” Palacios says. Palacios first earned a grant from the Deshpande Center for Technological Innovation to launch CEI. Afterward, he took his idea for GaN-based power electronics to Innovation Teams (i-Teams), which brings together MIT students from across disciplines to evaluate the commercial feasibility of new technologies. That program, he says, showed him the huge market pull for GaN power electronics, and helped CEI settle on its first products. “Many times, it’s the other way around: You come out with an amazing technology looking for an application. In this case, thanks to i-Teams, we found there were many applications looking for this technology,” Palacios says. For Lu, a key element for growing CEI was auditing Start6, a workshop hosted by the Department of Electrical Engineering and Computer Science, where entrepreneurial engineering students are guided through the startup process with group discussions and talks from seasoned entrepreneurs. Among other things, Lu gained perspective on dividing equity, funding, building a team, and other early startup challenges. “It’s a great class for a student who has an idea, but doesn’t know exactly what’s going on in business,” Lu says. “It’s kind of an overview of what the process is going to be like, so when you start your own company you are ready.”
Ian A. Waitz, dean of the School of Engineering, officially welcomed 183 students into the 2015-2016 SuperUROP class on Thursday, Sept. 30, at a Stata Center reception for participants in the program, which has more than doubled in size since its 2012 launch within the Department of Electrical Engineering and Computer Science (EECS). Opened for the first time this year to students throughout the School of Engineering, the Advanced Undergraduate Research Opportunities Program — commonly known as SuperUROP — is a yearlong independent research program that builds on the success of MIT’s Undergraduate Research Opportunities Program (UROP). SuperUROP’s additional goals include requiring students to take a two-term course on undergraduate research and to produce prototypes or publication-worthy results. “SuperUROP is a great opportunity to go into depth on a topic over a year and have some structure,” Waitz told this year’s new SuperUROP students. “I hope you have fun with it — I know the people before you did.” Waitz, the Jerome C. Hunsaker Professor of Aeronautics and Astronautics, said the School of Engineering provided support for the program’s expansion this year because the numbers of students participating through EECS showed a high level of interest. “Students have been voting with their feet,” he said after the event. “The participation was already there, and when you talked to the ones who had done it, they really loved the experience it gave them.” SuperUROP has grown every year since it began with 86 participants, all of whom were in EECS. This year, 183 students have signed up from seven departments: Aeronautics and Astronautics (AeroAstro); Biological Engineering; Chemical Engineering; Civil and Environmental Engineering; EECS; Mechanical Engineering; and Nuclear Science and Engineering. “The expansion of the SuperUROP program across the School of Engineering will enable our students to learn about a broad range of research directions,” said Anantha P. Chandrakasan, the Joseph F. and Nancy P. Keithley Professor of Electrical Engineering and head of EECS. “This year’s class is our biggest yet, and they are already making significant progress in formulating their research projects and technical approaches. I look forward to following their impactful research over the academic year.” Students attending the reception described working on a wide range of projects from machine learning to electrospray micro-thrusters and from cybersecurity to concussion-prevention football helmets. “SuperUROP is about really making a contribution to what’s already out there, which is super cool,” said Skanda Koppula, a junior in EECS who is trying to develop a chip capable of speaker authentication. “There’s a lot of emphasis on the project being pioneering in the field.” Designed for students who have already completed a UROP and who want to advance to graduate-level research, SuperUROP requires more independence — which students say they like. “You’re not a code monkey or a lab lackey. You are calling the shots with your own project,” said Kaustav Gopinathan, a senior in EECS who participated in the SuperUROP program last year and offered remarks at the reception. “I feel like we’re treated as grad students. There are tons of resources and plenty of access to the professor,” said Julia CrowleyFarenga, a senior in AeroAstro who is working on a football helmet. Focusing on a project for a full year is a major advantage of the SuperUROP, students said. “You don’t have to rush,” explained Berj Chilingirian, a senior in EECS and mathematics working on a hardware extension for improved computer security. “Really technical problems require a lot of time.” SuperUROP students aren’t expected to be entirely independent, however. The program provides a structure through which students can set, and accomplish, their research goals. Notably, they take 6.UAR (Seminar in Undergraduate Advanced Research), which covers such topics as choosing a research topic, industry best practices, communication skills, and ethics. “It’s helped me learn to present my research, which I hadn’t done before,” said Michael Burton, a senior in AeroAstro. “The class is really cool. They bring in speakers every Thursday,” said Harini Suresh, senior in EECS, noting that Sangeeta Bhatia, the John and Dorothy Wilson Professor at MIT’s Institute for Medical Engineering and Science and Electrical Engineering and Computer Science, recently explained her work on human microlivers to the class. Such presentations give students exposure to a wide range of research applications. “I think [SuperUROP is] a cool way to explore [… ] what you’re interested in and not interested in,” Suresh added. In sum, the program requires a lot of work, but the effort is worthwhile, according to Gopinathan. “SuperUROP is where you learn to tackle real research problems. It’s something you can’t get from classes,” he said. “SuperUROP was one of the most rewarding experiences I’ve had at MIT.” SuperUROP is made possible through the contributions of 19 industry sponsors and support from 13 alumni and friends of EECS.
Gao R.,Liverpool John Moores University |
Ji Z.,Liverpool John Moores University |
Zhang J.F.,Liverpool John Moores University |
Zhang W.D.,Liverpool John Moores University |
And 6 more authors.
IEEE Transactions on Semiconductor Manufacturing | Year: 2015
Characterizing positive charges and its energy distribution in gate dielectric is useful for process qualification. A discharge-based technique is introduced to extract their energy distribution both within and beyond substrate band-gap. This paper investigates the difficulties in its implementation on typical industrial parameter analyzer and provides solutions. For the first time, we demonstrate the technique's applicability to the advanced 22-nm fabrication process and its capability in evaluating the impact of different strains on the energy distribution. The test time is within several hours. This, together with its implementation on industrial parameter analyzer, makes it a useful tool in the semiconductor manufacturing foundries for process monitoring and optimization. © 2015 IEEE. Source