Evelyn Reynolds

In America’s quest to slash greenhouse gas emissions, many have cited the chemical industry as one of the hardest to decarbonize. It’s a significant roadblock: Chemical separation alone is responsible for up to 15 percent of the U.S.’s total energy usage.

Osmoses, a startup trying to dramatically increase the efficiency of chemical separations, got a major boost Thursday when it won the MIT $100K Entrepreneurship Competition. The company has developed a molecular filtration solution containing tiny channels that can be precisely sized to separate even the smallest molecules. The company claims its membranes can form channels that are 1/100,000 the width of a human hair, allowing the separation of molecules that differ in size by a mere fraction of an angstrom — less than the size of an atom.

“This is one of the greatest challenges of the century for our society, but also one of the biggest opportunities for companies that can innovate in this space,” Francesco Maria Benedetti, a postdoc at MIT, said in the winning pitch. The company is also led by PhD candidate Katherine Mizrahi Rodriguez ’17, Zachary P. Smith, the Joseph R. Mares Career Development Professor of Chemical Engineering at MIT, and Holden Lai, a postdoc at the University of Pennsylvania and former researcher in Smith’s lab.

Many chemical separation processes, such as distillation, use huge amounts of energy in the form of heat. Membrane filtration offers a promising alternative form of separation, but Osmoses says most membranes today have poor performance, leading to low adoption rates among chemical plants and higher operating costs.

Osmoses’ membranes come in a module that fits in existing separation systems. The company has tested its lab prototype in industrial-like environments, including in high-pressure, variable temperature conditions. The company says its results show a marked improvement over existing membrane filtration technologies.

“We’ve completely redesigned the materials these membranes are made of, lowering the energy consumption to a minimum and generating unprecedented performance,” Benedetti said.

The company is starting by targeting gas and vapor separations in the traditional and renewable natural gas processing space. Osmoses says by switching to its solution, companies in the market can reduce product loss by 85 percent, generating added fuel that could power 7 million additional homes in the U.S. for a year.

Osmoses also believes it can bring efficiencies to oxygen and nitrogen generation, hydrogen purification, and carbon capture.

The company will use the prize money to purchase equipment and scale its prototype later this year. Next year, it hopes to test an early version of its product with potential customers.

Osmoses’ first customers will be natural gas plants that produce hundreds of millions of standard cubic feet of gas per day. The team believes it can reduce up to 1 million tons of carbon dioxide emissions from each plant of that size.

The MIT $100K is MIT’s largest entrepreneurship competition. It began in 1989 (with a much smaller grand prize value) and is organized by students with support from the Martin Trust Center for MIT Entrepreneurship and the MIT Sloan School of Management. Each team must include at least one current MIT student.

The second place, $25,000 prize went to Mach 9, which is building a suite of tools to help companies locate and analyze underground utilities.

“We’re helping you look at what’s underground right now by creating the Google Maps for subsurface information,” said CEO Alex Baikovitz. “We’re making subsurface mapping as easy as driving a car through the city on a nice summer day — not like right after a Red Sox game.”

Baikovitz says many companies currently rely on painted lines to locate underground utilities, an error-prone solution that leads to billions of dollars in losses.

Mach 9 is developing a digital visualization tool to help utility locators interpret field data and save their results in the cloud for future reference. The solution automatically interprets radar data in real-time without an internet connection. It is also building a complementary tool to simplify ground surveys for large-scale construction projects. The software integrates with commonly used design and mapping tools in the industry.

“We’re creating geospatial postprocessing software that just makes sense and is intuitive,” Baikovitz said. “We create 3-D models of the subsurface environment, where we can overlay ground penetrating radar data for interpretation. We can show a digital twin of utilities identified by the Mach 9 system.”

Mach 9 is already in negotiations with the largest utility locating company in the U.S. and is collecting over 1,000 miles of ground-penetrating radar data from surveying firms. The company has also already acquired over $250,000 worth of mapping equipment to construct a proof of concept. It plans to make its first sales in 2022.

In the future, the Mach 9 team plans to offer underground mapping solutions in agriculture and mining. It estimates subsurface mapping to be a $90 billion industry.

This year’s event was hosted by Carly Chase, a senior lecturer at the MIT Sloan School of Management, and Scott Stern, the David Sarnoff Professor of Management and chair of the Technological Innovation, Entrepreneurship, and Strategic Management Group at the MIT Sloan School of Management. It also featured an interview with Payal Kadakia ’05, the CEO and founder of exercise scheduling platform ClassPass.

The competition was the culmination of a process that began in the winter with more than 80 applicants from all five of MIT’s schools pitching their ideas. Thursday’s winning teams were two of eight finalists. The other finalist teams were:

Azeki Road, which is building technology solutions to help consumer brands in Africa scale around the world;

Candelytics, an analytics company building solutions to make 3-D data created by technologies like LIDAR sensors more accessible, intelligent, and impactful;

UltraNeuro, which is building a wearable ultrasound transducer to activate damaged nerves and reduce pain caused by a condition called peripheral neuropathy;

Resolute, which is creating a line of natural sunscreens that is suited for people of all skin tones;

Synthera Health, which is developing a testing, analytics, and iron supplement platform to help people maintain optimal iron levels in their blood; and

Volt, which is creating a marketplace for companies to buy and sell printed circuit boards, reducing procurement times and costs. More

When Ben Calhoun and Dave Wentzloff co-founded Everactive in 2012, analysts and tech companies were forecasting a massive increase in the number of internet-connected devices, collectively referred to as the internet of things (IoT). IBM, for example, predicted a staggering 1 trillion IoT-connected devices by 2015.

But Calhoun and Wentzloff knew better. The pair, who’d met as graduate students researching ultra-low-power circuits in Anantha Chandrakasan’s research group at MIT, recognized that 1 trillion devices meant the near-impossible task of managing 1 trillion batteries to sustain all of the sensors needed to continuously collect, analyze, and send data.

“Ben and I recognized that there was no way those IoT projections could happen if all of the devices had to run on batteries,” says Wentzloff. “We started Everactive with the collective vision of ridding the IoT world of batteries while ushering in the next era of self-powered computing systems.”

Within an industrial environment like a manufacturing plant, there are often thousands of assets that require careful attention to ensure efficient operation. An effective real-time monitoring solution must continuously stream data from each of these points to keep information flowing in a timely fashion so that stakeholders can react quickly to any issues that might interrupt operation, damage equipment, or cause safety risks and environmental harm.

But this is tough to do at scale. Existing solutions that rely on wired sensing are cost-prohibitive to deploy at all of the locations necessary to make them useful. The same is true for battery-powered solutions, especially considering our collective struggle with short battery life.

“We’ve had customers in the industrial manufacturing context say they currently have a maintenance problem, but if they employ battery-powered solutions they’re just replacing their existing maintenance problem with a different maintenance problem of constantly replacing batteries across thousands of devices,” says Calhoun.

Now imagine that issue spread across a world where we have 1 trillion IoT devices. Even if we do get to a point where we have an IoT battery with a 10-year lifespan (the current industry goal), we’d be looking at changing more than 270 million batteries every day.

In addition to battery limitations, the scaling issues are compounded by existing wireless networking technology. To effectively address the needs of a manufacturing plant — let alone a 1 trillion-node world — we need extremely high-density sensor networks to communicate at relatively long range throughout environments that are notoriously inhospitable to wireless signals.

Everactive’s innovation solves both of these problems, simultaneously getting rid of the battery and reinventing low-power wireless networking with its differentiating technology: ultra-low-power integrated circuits.

Yes, the same technology Calhoun and Wentzloff researched as graduate students at the Institute. The same field they continued to work on as professors at their undergraduate alma maters, the University of Virginia and the University of Michigan, respectively. In fact, after completing their doctorates at MIT — Calhoun in 2006 and Wentzloff in 2007 — they maintained close ties and eventually began to cooperate across those university boundaries.

Calhoun’s research group focused on low-power digital systems, and Wentzloff’s group explored low-power communication. “Our specific areas of interest dovetail quite nicely to form complete solutions, so our groups started to collaborate — it became a very close connection,” explains Wentzloff.

When the two technical co-founders looked to expand their startup, they tapped a collection of their newly minted PhD students who had the expertise of developing wireless system-on-chip technologies in the lab. Today, Everactive has expanded into a team of nearly 90 industry veterans and technical experts, including talented minds like Alice Wang, who joined up with Calhoun and Wentzloff in 2018 after successful stints with industry giants Texas Instruments and MediaTek. Another MIT alum, she now serves as VP of hardware for Everactive, directing both silicon and hardware systems design.

“We’re exceptionally proud of the team that we’ve developed,” says Wentzloff. “I think a large part of why we continue to succeed is that we’ve done a great job of surrounding our core technology students with a broad set of talented industry leaders.” 

Thanks to their advances in ultra-low-power circuits and wireless communication, Everactive sells full-stack industrial IoT solutions powered by their always-on Eversensors, harvesting energy exclusively from the surrounding environment. The sensors can be deployed at a larger scale than battery-powered devices, and they cost less to operate. At the other end of their system, the Evercloud transforms new data into high-value, actionable insights.

Their first product is a steam trap monitor. In a small factory, there might be hundreds of these mechanisms, while large oil and chemical refineries are often home to thousands of them.

Steam may be an old technology, but it’s ubiquitous from an energy standpoint. We still use it in our power plants to generate electricity and it’s a common source of heat for our homes and buildings.

“Steam is a great way to move energy around a large area, which is why it’s still used across so many application verticals and marketing segments,” explains Calhoun. In a steam distribution network, as energy is extracted from the steam it turns back into water. Steam traps function as valves that allow the condensed water to exit the system while keeping the steam inside.

Their failure can lead to wasted energy, costly downtime if the process has to be taken offline, or even dangerous explosions. And monitoring thousands of steam traps for malfunctions is not that simple. “Some steam traps are buried in the ground, and some of them are located two or three stories up,” says Wentzloff. “It can be difficult to reach them. Hence, the manual inspection problem.”

But Everactive’s solution can be deployed on every single steam trap in a steam distribution network for real-time assessment of the state of the system. This is why top industry players like Armstrong International, the second-largest steam trap manufacturer globally, has chosen to partner with them. “Armstrong was so impressed with what we have to offer that we’re now working together to bring Everactive’s monitoring services to market, in conjunction with Armstrong’s excellent steam traps,” says Calhoun.

Successful case studies abound, which is a key factor in their being named to MIT STEX25. Just last year they deployed their sensors in an industrial facility with 1,200 steam traps. They saved that customer close to $2.5 million in energy costs and around 34,000 metric tons of CO2, “which is equivalent to about 7,000 passenger cars per year being taken off the road,” says Wentzloff. They also averted the loss of 60 million gallons of wastewater that would otherwise be leaking through failed traps.

But the industrial space isn’t the beginning and the end for Everactive. Consumer electronics, particularly wearables, is often talked about when discussing the IoT, and it’s very much on Everactive’s radar. Logistics is another area they are exploring. By adding localization technology to their next generation of sensors, Everactive could help provide the ability to track assets anywhere across the globe. “There’s an unbounded list of industries and applications where we can apply this technology. We’re really excited about what’s coming ahead,” says Wentzloff.

In the face of the coronavirus pandemic, businesses are reevaluating how they operate and how they will function going forward. With workforces reduced, especially in factory settings where Everactive’s clients are building products essential to the economy, it has become even more important to understand what’s going on in places that aren’t easily accessible.

Intent on filling that need, Everactive expects to play an essential role in driving self-powered solutions to make ubiquitous remote monitoring possible. “At Everactive, we believe in a future where all of our environments and assets are monitored and smart, capable of providing information to computing systems to make our world more efficient, safer, and just better for the way we live,” says Calhoun. “We want to be a part of that, and we think our batteryless solution is the right path forward.” More

MIT has released an ambitious new plan for action to address the world’s accelerating climate crisis. The plan, titled “Fast Forward: MIT’s Climate Action Plan for the Decade,” includes a broad array of new initiatives and significant expansions of existing programs, to address the needs for new technologies, new policies, and new kinds of outreach to bring the Institute’s expertise to bear on this critical global issue.

As MIT President L. Rafael Reif and other senior leaders have written in a letter to the MIT community announcing the new plan, “Humanity must find affordable, equitable ways to bring every sector of the global economy to net-zero carbon emissions no later than 2050.” And in order to do that, “we must go as far as we can, as fast as we can, with the tools and methods we have now.” But that alone, they stress, will not be enough to meet that essential goal. Significant investments will also be needed to invent and deploy new tools, including technological breakthroughs, policy initiatives, and effective strategies for education and communication about this epochal challenge.

“Our approach is to build on what the MIT community does best — and then aspire for still more. Harnessing MIT’s long record as a leader in innovation, the plan’s driving force is a series of initiatives to ignite research on, and accelerate the deployment of, the technologies and policies that will produce the greatest impact on limiting global climate change,” says Vice President for Research Maria Zuber, who led the creation and implementation of MIT’s first climate action plan and oversaw the development of the new plan alongside Associate Provost Richard Lester and School of Engineering Dean Anantha Chandrakasan.

The new plan includes a commitment to investigate the essential dynamics of global warming and its impacts, increasing efforts toward more precise predictions, and advocating for science-based climate policies and increased funding for climate research. It also aims to foster innovation through new research grants, faculty hiring policies, and student fellowship opportunities.

Decarbonizing the world’s economy in time will require “new ideas, transformed into practical solutions, in record time,” the plan states, and so it includes a push for research focused on key areas such as cement and steel production, heavy transportation, and ways to remove carbon from the air. The plan affirms the imperative for decarbonization efforts to emphasize the need for equity and fairness, and for broad outreach to all segments of society.

Charting a shared course for the future

Having made substantial progress in implementing the Institute’s original five-year Plan for Action on Climate Change, MIT’s new plan outlines measures to build upon and expand that progress over the next decade. The plan consists of five broad areas of action: sparking innovation, educating future generations, informing and leveraging government action, reducing MIT’s own climate impact, and uniting and coordinating all of MIT’s climate efforts.

MIT is already well on its way to reaching the initial target, set in 2015, to reduce the Institute’s net carbon emissions by at least 32 percent from 2005 levels by the year 2030. That goal is being met through a combination of innovative off-campus power purchase agreements that enable the construction of large-scale solar and wind farms, and an array of renewable energy and building efficiency measures on campus. In the new plan, MIT commits to net-zero direct carbon emissions by 2026.

The initial plan focused largely on intensifying efforts to find breakthrough solutions for addressing climate change, through a series of actions including the creation of new low-carbon energy centers for research, and the convening of researchers, industry leaders, and policymakers to facilitate the sharing of best practices and successful measures. The new plan expands upon these actions and incorporates new measures, such as climate-focused faculty positions and student work opportunities to help tackle climate issues from a variety of disciplines and perspectives.

A long-running series of symposia, community forums, and other events and discussions helped shape a set of underlying principles that apply to all of the plan’s many component parts. These themes are:

The centrality of science, to build on MIT’s pioneering work in understanding the dynamics of global warming and its effects;
The need to innovate and scale, requiring new ideas to be made into practical solutions quickly;
The imperative of justice, since many of those who will be most affected by climate change are among those with the least resources to adapt;
The need for engagement, dealing with government, industry, and society as a whole, reflecting the fact that decarbonizing the world’s economy will require working with leaders in all sectors; and
The power of coordination, emphasizing the need for the many different parts of the Institute’s climate research, education, and outreach to have clear structures for decision making, action, and accountability.

Bolstering research and innovation

The new plan features a wide array of action items to encourage innovation in critical areas, including new programs as well as the expansions of existing programs. This includes the Climate Grand Challenges, announced last year, which focus on game-changing research advances across disciplines spanning MIT.

“We must, and we do, call for critical self-examination of our own footprint, and aspire for substantial reductions. We also must, and we do, renew and bolster our commitment to the kind of paradigm-shifting research and innovation, across every sector imaginable (and some perhaps still waiting to be discovered), that the world expects from MIT,” Lester says. “An immediate and existential crisis like climate change calls for both near-term and extraordinary long-term strokes. I believe the people of MIT are capable of both.” 

The plan also calls for expanding the MIT Climate and Sustainability Consortium, created earlier this year, to foster collaborations among companies and researchers to work for solutions to climate problems. The aim is to greatly accelerate the adoption of large-scale, real-world climate solutions, across different industries around the world, by working with large companies as they work to find ways to meet new net-zero climate targets, in areas ranging from aerospace to packaged food.

Another planned action is to establish a Future Energy Systems Center, which will coalesce the work that has been fostered through MIT’s Low-Carbon Energy Centers, created under the previous climate action plan. The Institute is also committing to devoting at least 20 upcoming faculty positions to climate-focused talent. And, there will be new midcareer ignition grants for faculty to spur work related to climate change and clean energy.

For students, the plan will provide up to 100 new Climate and Sustainability Energy Fellowships, spanning the Institute’s five schools and one college. These will enable work on current or new projects related to climate change. There will also be a new Climate Education Task Force to evaluate current offerings and make recommendations for strengthening research on climate-related topics. And, in-depth climate or clean-energy-related research opportunities will be offered to every undergraduate who wants one. Climate and sustainability topics and examples will be introduced into courses throughout the Institute, especially in the General Institute Requirements that all undergraduates must take.

This emphasis on MIT’s students is reflected in the plan’s introductory cover letter from Reif, Zuber, Lester, Chandrakasan, and Executive Vice President and Treasurer Glen Shor. They write: “In facing this challenge, we have very high expectations for our students; we expect them to help make the impossible possible. And we owe it to them to face this crisis by coming together in a whole-of-MIT effort — deliberately, wholeheartedly, and as fast as we can.”

The plan’s educational components provide “the opportunity to fundamentally change how we have our graduates think in terms of a sustainable future,” Chandrakasan says. “I think the opportunity to embed this notion of sustainability into every class, to think about design for sustainability, is a very important aspect of what we’re doing. And, this plan could significantly increase the faculty focused on this critical area in the next several years. The potential impact of that is tremendous.”

Reaching outward

The plan calls for creating a new Sustainability Policy Hub for undergraduates and graduate students to foster interactions with sustainability policymakers and faculty, including facilitating climate policy internships in Washington. There will be an expansion of the Council on the Uncertain Human Future, which started last year to bring together various groups to consider the climate crisis and its impacts on how people might live now and in the future.

“The proposed new Sustainability Policy Hub, coordinated by the Technology and Policy Program, will help MIT students and researchers engage with decision makers on topics that directly affect people and their well-being today and in the future,” says Noelle Selin, an associate professor in the Institute for Data, Systems, and Society and the Department of Earth, Atmospheric, and Planetary Sciences. “Ensuring sustainability in a changed climate is a collaborative effort, and working with policymakers and communities will be critical to ensure our research leads to action.”

A new series of Climate Action Symposia, similar to a successful series held in 2019-2020, will be convened. These events may include a focus on climate challenges for the developing world. In addition, MIT will develop a science- and fact-based curriculum on climate issues for high school students. These will be aimed at underserved populations and at countering sources of misinformation.

Building on its ongoing efforts to provide reliable, evidence-based information on climate science, technology, and policy solutions to policymakers at all levels of government, MIT is establishing a faculty-led Climate Policy Working Group, which will work with the Institute’s Washington office to help connect faculty members doing relevant research with officials working in those areas.

In the financial arena, MIT will lead more research and discussions aimed at strengthening the financial disclosures relating to climate that corporations need to make, thus making the markets more sensitive to the true risks to investors posed by climate change. In addition, MIT will develop a series of case studies of companies that have made a conversion to decarbonized energy and to sustainable practices, in order to provide useful models for others.

MIT will also expand the reach of its tools for modeling the impacts of various policy decisions on climate outcomes, economics, and energy systems. And, it will continue to send delegations to the major climate policy forums such as the UN’s Conference of the Parties, and to find new audiences for its Climate Portal, web-based Climate Primer, and TILclimate podcast.

“This plan reaffirms MIT’s commitment to developing climate change solutions,” says Christopher Knittel, the George P. Shultz Professor of Applied Economics. “It understands that solving climate change will require not only new technologies but also new climate leaders and new policy. The plan leverages MIT’s strength across all three of these, as well as its most prized resources: its students. I look forward to working with our students and policymakers in using the tools of economics to provide the research needed for evidence-based policymaking.”

Recognizing that the impacts of climate change fall most heavily on some populations that have contributed little to the problem but have limited means to make the needed changes, the plan emphasizes the importance of addressing the socioeconomic challenges posed by major transitions in energy systems, and will focus on job creation and community support in these regions, both domestically and in the developing world. These programs include the Environmental Solutions Initiative’s Natural Climate Solutions Program, and the Climate Resilience Early Warning System Network, which aims to provide fine-grained climate predictions.

“I’m extraordinarily excited about the plan,” says Professor John Fernández, director of the Environmental Solutions Initiative and a professor of building technology. “These are exactly the right things for MIT to be doing, and they align well with an increasing appetite across our community. We have extensive expertise at MIT to contribute to diverse solutions, but our reach should be expanded and I think this plan will help us do that.”

“It’s so encouraging to see environmental justice issues and community collaborations centered in the new climate action plan,” says Amy Moran-Thomas, the Alfred Henry and Jean Morrison Hayes Career Development Associate Professor of Anthropology. “This is a vital step forward. MIT’s policy responses and climate technology design can be so much more significant in their reach with these engagements done in a meaningful way.”

Decarbonizing campus

MIT’s first climate action plan produced mechanisms and actions that have led to significant reductions in net emissions. For example, through an innovative collaborative power purchase agreement, MIT enabled the construction of a large solar farm and the early retirement of a coal plant, and also provided a model that others have since adopted. Because of the existing agreement, MIT has already reduced its net emissions by 24 percent despite a boom in construction of new buildings on campus. This model will be extended moving forward, as MIT explores a variety of possible large-scale collaborative agreements to enable solar energy, wind energy, energy storage, and other emissions-curbing facilities.

Using the campus as a living testbed, the Institute has studied every aspect of its operations to assess their climate impacts, including heating and cooling, electricity, lighting, materials, and transportation. The studies confirm the difficulties inherent in transforming large existing infrastructure, but all feasible reductions in emissions are being pursued. Among them: All new purchases of light vehicles will be zero-emissions if available. The amount of solar generation on campus will increase fivefold, from 100 to 500 megawatts. Shuttle buses will begin converting to electric power no later than  2026, and the number of car-charging stations will triple, to 360.

Meanwhile, a new working group will study possibilities for further reductions of on-campus emissions, including indirect emissions encompassed in the UN’s Scope 3 category, such as embedded energy in construction materials, as well as possible measures to offset off-campus Institute-sponsored travel. The group will also study goals relating to food, water, and waste systems; develop a campus climate resilience plan; and expand the accounting of greenhouse gas emissions to include MIT’s facilities outside the campus. It will encourage all labs, departments, and centers to develop plans for sustainability and reductions in emissions.

“This is a broad and appropriately ambitious plan that reflects the headway we’ve made building up capacity over the last five years,” says Robert Armstrong, director of the MIT Energy Initiative. “To succeed we’ll need to continually integrate new understanding of climate science, science and technology innovations, and societal engagement from the many elements of this plan, and to be agile in adapting ongoing work accordingly.”

Examining investments

To help bring MIT’s investments in line with these climate goals, MIT has already begun the process of decarbonizing its portfolio, but aims to go further.

Beyond merely declaring an aspirational goal for such reductions, the Institute will take this on as a serious research question, by undertaking an intensive analysis of what it would mean to achieve net-zero carbon by 2050 in a broad investment portfolio.

“I am grateful to MITIMCO for their seriousness in affirming this step,” Zuber says. “We hope the outcome of this analysis will help not just our institution but possibly other institutional managers with a broad portfolio who aspire to a net-zero carbon goal.”

MIT’s investment management company will also review its environmental, social, and governance investment framework and post it online. And, as a member of Climate Action 100+, MIT will be actively engaging with major companies about their climate-change planning. For the planned development of the Volpe site in Kendall square, MIT will offset the entire carbon footprint and raise the site above the projected 2070 100-year flood level.

Institute-wide participation

A centerpiece of the new plan is the creation of two high-level committees representing all parts of the MIT community. The MIT Climate Steering Committee, a council of faculty and administrative leaders, will oversee and coordinate MIT’s strategies on climate change, from technology to policy. The steering committee will serve as an “orchestra conductor,” coordinating with the heads of the various climate-related departments, labs, and centers, as well as issue-focused working groups, seeking input from across the Institute, setting priorities, committing resources, and communicating regularly on the progress of the climate plan’s implementation.

The second committee, called the Climate Nucleus, will include representatives of climate- and energy-focused departments, labs, and centers that have significant responsibilities under the climate plan, as well as the MIT Washington Office. It will have broad responsibility for overseeing the management and implementation of all elements of the plan, including program planning, budgeting and staffing, fundraising, external and internal engagement, and program-level accountability. The Nucleus will make recommendations to the Climate Steering Committee on a regular basis and report annually to the steering committee on progress under the plan.

“We heard loud and clear that MIT needed both a representative voice for all those pursuing research, education, and innovation to achieve our climate and sustainability goals, but also a body that’s nimble enough to move quickly and imbued with enough budgetary oversight and leadership authority to act decisively. With the steering committee and Climate Nucleus together, we hope to do both,” Lester says.

The new plan also calls for the creation of three working groups to address specific aspects of climate action. The working groups will include faculty, staff, students, and alumni and give these groups direct input into the ongoing implementation of MIT’s plans. The three groups will focus on climate education, climate policy, and MIT’s own carbon footprint. They will track progress under the plan and make recommendations to the Nucleus on ways of increasing MIT’s effectiveness and impact.

“MIT is in an extraordinary position to make a difference and to set a standard of climate leadership,” the plan’s cover letter says. “With this plan, we commit to a coordinated set of leadership actions to spur innovation, accelerate action, and deliver practical impact.”

“Successfully addressing the challenges posed by climate change will require breakthrough science, daring innovation, and practical solutions, the very trifecta that defines MIT research,” says Raffaele Ferrari, the Cecil and Ida Green Professor of Oceanography. “The MIT climate action plan lays out a comprehensive vision to bring the whole Institute together and address these challenges head on. “Last century, MIT helped put humans on the moon. This century, it is committing to help save humanity and the environment from climate change here on Earth.” More

These days homebuilders might have several reasons to make new homes energy-efficient. They may be required to hit efficiency goals by local building codes. They may want to take advantage of financial incentive programs offered by governments, lenders, and utilities. They may just want to appeal to the growing segment of home buyers who prioritize sustainability and want lower energy bills.

But the process of building energy-efficient homes and then getting certifications requires cooperation across a complex ecosystem of players. For the last 10 years, Ekotrope has worked to simplify that process.

The company’s software was inspired by system optimization work done for NASA by Ed Crawley, Ford Professor of Engineering at MIT and co-founder of Ekotrope. It brings together disparate systems used by builders, home energy raters, and utilities to calculate the efficiency and costs of different designs. Energy raters can then use Ekotrope’s system to apply for home energy certifications. If the criteria aren’t met, the system gives reasons why. If the submission is successful, Ekotrope completes the accreditation process instantaneously.

“The problem we are trying to solve is that information does not flow very well,” co-founder and CEO Ziv Rozenblum SM ’07 says. “For example, previously, if a builder wanted to participate in an energy efficiency program, they’d send a file, it could take months to get feedback, they’d make corrections, and many hands would touch that file. We automated almost everything.”

Today Ekotrope is one of the leading energy-accreditation systems in the country. The company says its software has been used to certify more than half a million homes and is used in the construction of one in every five new homes in the U.S.

The company’s success translates to major impact in a home energy sector responsible for a fifth of all U.S. greenhouse gas emissions. For the founders, the success affirms their belief that the U.S. can make huge strides in reducing carbon emissions using today’s technologies — as long as the right systems are in place.

“I see all this interest in inventing new technologies and building energy-efficient solutions, but we think a lot of progress can be achieved with existing solutions,” Rozenblum says. “You just need to help people make the right choices at the right time. All the stakeholders want to make the right decisions; they just don’t always have the right information.”

Building a better system

The idea for Ekotrope, like so many successful businesses, came from a bad experience. Crawley was working with a contractor and architect to build a new home and was disappointed that neither person could project the impact different materials and appliances would have on the overall energy efficiency of the home. At MIT, Crawley had worked with NASA and BP on projects in which researchers had to determine the impact of different parts on efficiency, performance, cost, and more.

“He had a light bulb go off that designing a home is a similarly complex process,” Ekotrope co-founder and lead engineer Nick Sisler ’11 says. “There are a lot of options. Specifically around energy efficiency, there are all these different components of a home that affect energy consumption and cost — whether it’s insulation, heating and cooling systems, solar panels on the roof, light bulbs — all of those things have an impact on energy and cost.”

In 2010, Cy Kilbourn, a visiting researcher at MIT from Brown University, and Rozenblum, who had been a research assistant for Crawley as a graduate student, worked with Crawley to understand how different home construction decisions impacted energy efficiency. The following year Rozenblum quit his job to run Ekotrope full time. Sisler, who had researched the home energy preferences of buyers with Crawley as an undergraduate, joined shortly after graduation in 2011. The other founders are software engineer Ben DeLillo and Kenneth Lazarus SM ’89, PhD ’92.

The founders initially began building a software solution for architects and builders, calculating the costs associated with different design options and their impact on efficiency and emissions. A key component of the solution was an algorithm that measured hourly energy use in different scenarios.

Around 2016, Ekotrope pivoted to selling to home energy raters. Raters sit at the heart of home energy accreditations, working with builders, utilities, accreditation agencies, mortgage lenders, and governments, and providing an energy score for climate-conscious buyers.

“The [energy raters] will work with the builder, get their building plans, put that data into Ekotrope, and see what energy consumption is predicted to be, what energy codes the home will need, what programs it qualifies for, like Energy Star or tax credit or utility rebate programs,” Sisler explains. “All that stuff is integrated into our solution.”

The system streamlines a process the founders say had prevented energy efficiency programs from reaching their full potential.

“People are making worse decisions because they lack information, and there’s lot of double data entry and inefficiencies,” Rozenblum says. “We try to solve that by making systems that provide people with the information they need to make better choices.”

Leaving a large footprint

The founders say about 75 percent of new energy-efficient homes in the U.S. are accredited with help from Ekotrope’s software. Most of those homes are single-family.

The company has partnered with some of the largest programs promoting home energy efficiency in the country. Ekotrope has also partnered with mortgage lenders, material suppliers, and about 40 utilities.

That progress has put Ekotrope in a unique position to help different players in the industry understand what kind of incentives improve sustainability and what other trends they need to prepare for.

“We probably have the most inclusive database of information on new homes,” Rozenblum says. “We have information like who’s building new homes, where, what kind of materials they’re using, how far they are from an energy goal, how much CO2 they will add to the atmosphere, what’s the projected performance, what kind of incentives are working and not working.”

Ekotrope also sees opportunities to work more closely with utilities, and has seen strong results from pilot programs that let utilities make suggestions to raters and builders.

“It’s exciting to show that energy efficiency and economic decisions aren’t different,” Rozenblum says. “You can make money and be efficient at the same time.” More

The Commonwealth of Massachusetts recently passed a climate bill that sets a target of net-zero emissions for the state by the year 2050. The bill is one of several successful legislative efforts in Northeastern states to reduce greenhouse gas emissions by as much as 80 to 100 percent by mid-century. To achieve these ambitious targets — which align with the Paris Agreement’s long-term goal of keeping global warming well below 2 degrees Celsius to avoid the worst impacts of climate change — will require a significant ramp-up of zero-carbon, intermittent, renewable energy technologies.

Hydropower is a particularly appealing renewable energy option for policymakers in the region; substantial hydro resources available in nearby Quebec could be used to dispatch power to consumers in Northeastern states during periods of low wind and solar generation. But environmental and aesthetic concerns have mobilized communities along proposed hydro transmission line routes to nip that notion in the bud. To stand a chance of overcoming these concerns, policymakers in the U.S. Northeast and Quebec will need to demonstrate compelling benefits to consumers and transmission line abutters alike.

To that end, researchers at the MIT Joint Program on the Science and Policy of Global Change and MIT Energy Initiative have conducted a study to assess the economic impacts of expanding hydropower transmission capacity from Quebec to the Northeast. Using a unique modeling framework that represents both regional economic behavior and hourly electricity operations, they project these impacts under three scenarios. In each scenario, transmission capacity is expanded by 10, 30, or 50 percent above existing capacity into New York and all New England states starting in 2026, and carbon emissions are capped in alignment with regional climate goals.

Compared to a reference scenario in which current and projected state renewable energy technology policies are implemented with carbon emissions capped to achieve mid-century regional goals, the researchers estimate that by 2050, electricity imports enabled by these three transmission expansions save the New York state economy 38-40 cents per kilowatt hour (KWh) and the New England economy 30-33 cents per kWh. The results appear in the journal Energy Policy.

“These economy-wide savings are significantly higher than the cost of the electricity itself,” says Joint Program research scientist Mei Yuan, the lead author of the study. “Moreover, the carbon limits that we impose in these scenarios raise fuel prices enough to make electricity cost-competitive in multiple economic sectors. This accelerates electrification in both New England and New York, particularly between 2030 and 2050.”

The overall economic impact of the three transmission capacity expansion scenarios is a significantly lower cost of meeting the emissions reduction goals of all states in the region.

The study is an outgrowth of an Energy Modeling Forum effort, EMF34, which aims to improve understanding of how energy markets affect one another throughout North America. The researchers were supported by sponsors of the MIT Joint Program sponsors and the MIT Energy Initiative Seed Fund Program. More

“There is no lone genius who solves all the problems.”

Dennis Whyte, director of the Plasma Science and Fusion Center (PSFC), is reflecting on a guiding belief behind his nuclear science and engineering class 22.63 (Principles of Fusion Engineering). He has recently watched his students, working in teams, make their final presentations on how to use fusion technology to create carbon-free fuel for shipping vessels. Since taking on the course over a decade ago, Whyte has moved away from standard lectures, prodding the class to work collectively on finding solutions to “real-world” issues. Over the past years the course, and its collaborative approach to design, has been instrumental in guiding the real future of fusion at the PSFC.

For decades researchers have explored fusion, the reaction that powers the sun, as a potential source of virtually endless, carbon-free energy on Earth. MIT has studied the process with a series of “Alcator” tokamaks, compact machines that use high magnetic fields to keep the hot plasma inside and away from the walls of a donut-shaped vacuum vessel long enough for fusion to occur. But understanding how plasma affects tokamak materials, and making the plasma dense and hot enough to sustain fusion reactions, has been elusive.

Incubating fusion machines and design teams

The second time he taught the course, Whyte was ready for his students to attack problems related to net-energy tokamak operation, necessary to produce substantial and economical power. These problems could not be explored with the PSFC’s Alcator C-Mod tokamak, which maintained fusion in only brief pulses, but they could be studied by a class tasked with designing a fusion device that can operate around the clock.

Around this time Whyte learned of high-temperature superconducting (HTS) tape, a newly available class of superconducting material that supported creating higher magnetic fields for effectively confining the plasma. It had the potential to surpass the performance of the previous generation of superconductors, like niobium-tin, which was being used in ITER, the burning plasma fusion experiment being built in France. Could the class design a machine that would answer questions about steady-state operation, while taking advantage of this revolutionary product? Furthermore, what if components of the machine could be easily taken out and replaced or altered, making the tokamak flexible for different experiments?

What the class conceived was a tokamak called “Vulcan.” Whyte calls his students’ efforts “eye-opening,” original enough to produce five peer-reviewed articles for Fusion Engineering and Design. Although the tokamak design was never directly built, its exploration of demountable magnetic coils, made from the new HTS tape, suggested a path for a fusion future.

Two years later, Whyte started his students down that path. He asked, “What would happen in a device where we try to make 500 megawatts of fusion power — identical to what ITER does — but we use this new HTS technology?”

With student teams working on separate aspects of the project and coordinating with other groups to create an integrated design, Whyte decided to make the class environment even more collaborative. He invited PSFC fusion experts to contribute. In this “collective community teaching” environment the students expanded on the research from the previous class, creating the basis for HTS magnets and demountable coils.

As before, the innovations explored resulted in a published paper. The lead author was then-graduate student Brandon Sorbom PhD ’17. He introduced the fusion community to ARC, describe in the article’s title as “a compact, high-field, fusion nuclear science facility and demonstration power plant with demountable magnets.” Because ARC was too large a project to consider building immediately, Whyte and some of his postdocs and students eventually began thinking about how they could study the most important elements of the ARC design in a smaller device.

Their answer was SPARC, based on the experience gained from designing Vulcan and ARC. This compact, high-field, net fusion energy experiment has become a collaboration between MIT and Commonwealth Fusion Systems (CFS), a Cambridge, Massachusetts-based startup seeded with talent from 22.63. Bob Mumgaard and Dan Brunner, who helped design Vulcan, are in CFS leadership, as is Brandon Sorbom. MIT NSE Assistant Professor Zach Hartwig, who participated as a student in the Vulcan project, has also stayed involved in the SPARC project and developments. 

The economic question

The course had become an incubator for researchers interested in using the latest technology to re-imagine how quickly a fusion power plant would be possible. It helped redirect the focus of the PSFC from Alcator C-Mod, which ended operation in 2016, toward SPARC and ARC, and technology innovation. In the process the PSFC, whose fusion program had been largely funded by the U.S. Department of Energy, realized it would also need to expand its research sponsorship to private funding.

The discussions with the private sector brought home the requirement not just for technical feasibility, but for making fusion an attractive product economically. This inspired Whyte to add an economic constraint to the 2020 22.63 class project, noting “it changes how you think about attacking the design.” Consequently, he expanded the teaching team to include Eric Ingersoll, founder and managing director at LucidCatalyst and TerraPraxis. Together they imagined a novel application and market that could use fusion as an intense carbon-free energy source — international shipping.

The virtual nature of this year’s course offered the unique chance for a number of students, postdocs, and teachers from Princeton University to join the class as volunteers, with the intent of eventually creating a similarly structured course at Princeton. They integrated with MIT students and instructors into four teams working interdependently to design an onboard method of generating ammonia fuel for ship engines. The device was dubbed “ARCH,” the H standing for Hydrogen. By making innovations to the fusion design, mostly focused on improving materials and heat removal, the team showed they could meet economic targets.

For MIT graduate student Rachel Bielajew, part of the Systems Integration Team, focusing on the economics of the project provided a very different experience from her other classes and everyday research.

“It was definitely motivating to have an economic target driving design choices,” she says. “The class also reinforced for me that the pathway to successful fusion reactors is multidisciplinary and there is important research to be done in many fields.”

Whyte’s teaching journey has been as transformative for him as for his students.

“If you give young people the time, the tools, and the imaginative space to work together towards meaningful goals — it’s hard to imagine a more powerful force,” he says. “The class and the innovation provided by the collective student effort have changed my worldview, and, I believe, the prospects for fusion energy.” More

In recent decades, China’s rapid economic growth has enabled more and more consumers to buy their own cars. The result has been improved mobility and the largest automotive market in the world — but also serious urban air pollution, high greenhouse gas emissions, and growing dependence on oil imports.

To counteract those troubling trends, the Chinese government has imposed policies to encourage the adoption of plug-in electric vehicles (EVs). Since buying an EV costs more than buying a conventional internal combustion engine (ICE) vehicle, in 2009 the government began to provide generous subsidies for EV purchases. But the price differential and the number of buyers were both large, so paying for the subsidies became extremely costly for the government.

As a result, China’s policymakers planned to phase out the subsidies at the end of 2020 and instead impose a mandate on car manufacturers. Simply stated, the mandate requires that a certain percent of all vehicles sold by a manufacturer each year must be battery-powered. To avoid financial penalties, every year manufacturers must earn a stipulated number of points, which are awarded for each EV produced based on a complex formula that takes into account range, energy efficiency, performance, and more. The requirements get tougher over time, with a goal of having EVs make up 40 percent of all car sales by 2030.

This move will have a huge impact on the worldwide manufacture of EVs, according to William H. Green, the Hoyt C. Hottel Professor in Chemical Engineering. “This is one of the strongest mandates for electric cars worldwide, and it’s being imposed on the largest car market in the world,” he says. “There will be a gigantic increase in the manufacture of EVs and in the production of batteries for them, driving down the cost of both globally.”

But what will be the impact of the mandate within China? The transition to EVs will bring many environmental and other benefits. But how much will it cost the nation? In 2016, MIT chemical engineering colleagues Green and then-graduate student I-Yun Lisa Hsieh PhD ’20 decided to find out. Their goal was to examine the mixed impacts of the mandate on all affected factors: battery prices, manufacturing costs, vehicle prices and sales, and the cost to the consumer of owning and operating a car. Based on their results, they could estimate the total societal cost of complying with the mandate in the coming decade. (Note that the Chinese government recently extended subsidy support for EVs for two years due to the Covid-19 pandemic and that this analysis was performed before that change was announced.)

Looking at battery prices

“The main reason why EVs are costly is that their batteries are expensive,” says Green. In recent years, battery prices have dropped rapidly, largely due to the “learning effect”: As production volumes increase, manufacturers find ways to improve efficiency, and costs go down. It’s generally assumed that battery prices will continue to decrease as EVs take over more of the car market.

Using a new modeling approach, Green and Hsieh determined that learning effects will lower costs appreciably for battery production, but not much for the mining and synthesis of critical battery materials. They concluded that the price of the most widely used EV battery technology — the lithium-ion nickel-manganese-cobalt battery — will indeed drop as more are manufactured. But the decline will slow as the price gets closer to the cost of the raw materials in it.

Using the resulting estimates of battery price, the researchers calculated the extra cost of manufacturing an EV over time and — assuming a standard markup for profit — determined the likely selling price for those cars. In previous work, they had used a variety of data sources and analytical techniques to determine “affordability” for the Chinese population — in other words, the fraction of their income available to spend on buying a car. Based on those findings, they examined the expected impact on car sales in China between 2018 and 2030.

As a baseline for comparison, the researchers first assumed a “counterfactual” (not true-to-life) scenario — car sales without significant adoption of EVs, so without the new mandate. Under that assumption, annual projected car sales climb to more than 34 million by 2030.

When the subsidy on EV purchases is eliminated and the mandate is enacted in 2020, total car sales shrink. But thereafter, the growing economy and rising incomes increase consumer purchasing power and drive up the demand for private car ownership. Annual sales are on average 20 percent lower than in the counterfactual scenario, but they’re projected to reach about 30 million by 2030.

The researchers also projected the breakdown in sales between ICE vehicles and battery EVs at three points in time. According to that analysis, in 2020, EVs make up just 7 percent of the total (1.6 million vehicles). By 2025, that share is up to 21 percent (5.4 million). And by 2030, it’s up to 37 percent (11.2 million) — close to the government’s 40 percent target. Altogether, 66 million EVs are sold between 2020 and 2030.

Those results also track the split between two types of plug-in EVs: pure battery EVs and hybrid EVs (which are powered by both batteries and gasoline). About twice as many pure battery EVs are sold than hybrid EVs, even though the former are more expensive due to the higher cost of their batteries. “The mandate includes a special preference for cars with a longer range, which means cars with large batteries,” says Green. “So carmakers have a big incentive to manufacture the pure battery EVs and be awarded extra points under the mandate formula.”

For the consumer, the added cost of owning an EV includes any difference in vehicle expenses over the whole lifetime of the car. To calculate that difference, the researchers quantified the “total cost of ownership,” or TCO, including the purchase cost, fuel cost, and operating and maintenance costs (including insurance) of their two plug-in EVs and an ICE vehicle out to 2030.

Their results show that before 2020, owning either type of plug-in EV is less costly than owning an ICE vehicle due to the subsidy paid on EV purchases. After the subsidy is removed and the mandate imposed in 2020, owning a hybrid EV is comparable to owning an ICE vehicle. Owning a pure battery EV is more expensive due to its high-cost batteries. Dropping battery prices reduces total ownership cost for both types of EVs, but the pure battery EV remains more expensive out to 2030.

Cost to society

The next step for the researchers was to calculate the total cost to China of forcing the adoption of EVs. The basic approach is straightforward: They take the extra TCO for each EV sold in each year, discount that cost to its present value, and multiply the resulting figure by the number of cars sold in that year. (They exclude taxes embedded in the purchase prices of the vehicle, of electricity and gasoline, and so on, as the society will have to pay other taxes to replace that lost revenue.)

Using that methodology, they calculated the incremental cost to society of each EV sold in each year as well as the extra cost per kilometer driven, assuming that the vehicle has a lifetime of 12 years and is driven 12,500 kilometers each year. The results show that the incremental cost of owning and driving an EV decreases from 2021 to 2030. The cost declines more for pure battery EVs than for hybrid EVs, but the former remain more costly.

By combining the per-car cost to society with the number of cars sold, the researchers calculated the total extra cost incurred. In their results, the total number of EVs sold in a year more than offsets any decrease in per-vehicle cost, so the incremental cost to society grows. And that cost is sizeable. On average, the transition to EVs forced by the mandate will cost 100 billion yuan per year from 2021 to 2030, which is about 2 percent of the nationwide expenditure in the transport sector every year.

During the 10 years from 2021-30, the annual societal cost of the transition to almost 40 percent EVs is equivalent to about 0.1 percent of China’s growing gross domestic product. “So the cost to society of forcing the sale of EVs in place of ICE vehicles is significant,” says Hsieh. “People will have far less money in their pockets to spend on other purchases.”

Other considerations

Green and Hsieh stress that the high societal cost of the forced EV adoption must be considered in light of the potential benefits to be gained. For example, switching from ICE vehicles to EVs will lower air pollution and associated health costs; reduce carbon dioxide emissions to help mitigate climate change; and reduce reliance on imported petroleum, enhancing the country’s national energy security and balance of payments.

Hsieh is now working to quantify those benefits so that the team can perform a proper cost-benefit analysis of China’s transition to EVs. Her initial results suggest that the monetized benefits are — like the costs — substantial. “The benefits appear to be the same order of magnitude as the costs,” she says. “It’s so close that we need to be careful to get the numbers right.”

The researchers cite two other factors that may impact the cost side of the equation. In early 2018, six Chinese megacities with high air pollution began restricting the number of license plates issued for ICE vehicles and charging high fees for them. With their lower-cost, more-abundant “green car plates,” EVs became cost-competitive, and sales soared. To protect Chinese carmakers, the national government recently announced that it plans to end those restrictions. The outcome and its impacts on EV sales remain uncertain. (Again, due to the pandemic, policies restricting car ownership have mostly been relaxed for now.)

The second caveat concerns how carmakers price their vehicles. The results reported here assume that prices are calculated as they are today: the cost of manufacturing the vehicle plus a certain percentage markup for profit. With the new mandate in place, automakers will need to change their pricing strategy so as to persuade enough buyers to purchase EVs to reach the required fraction. “We don’t know what they’re going to do, but one possibility is that they’ll lower the price of their battery cars and raise the price of their gasoline cars,” says Green. “That way, they can still make their profits while operating within the law.” As an example, he cites how U.S. carmakers responded to Corporate Average Fuel Economy standards by adjusting the relative prices of their low- and high-efficiency vehicles.

While such a change in Chinese automakers’ pricing strategy would lower the price of EVs, it would also push up average car prices overall, because the total car sales mix is dominated by ICE vehicles. “Some people in China who would otherwise be able to afford a cheap gasoline car now won’t be able to afford it,” says Hsieh. “They’ll be priced out of the market.”

Green emphasizes the impact of the mandate on all carmakers worldwide. “I can’t overstate how hugely important this is,” he says. “As soon as the mandate came out, carmakers realized that electric vehicles had become a major market rather than a niche market on the side.” And he believes that even without subsidies, the added expense of buying an EV won’t be prohibitive for many car buyers — especially in light of the benefits they offer.

However, he does have a final concern. As more and more EVs are manufactured, global supplies of critical battery materials will become increasingly limited. At the same time, however, the supply of spent batteries will increase, creating an opportunity to recycle critical materials for use in new batteries and simultaneously prevent environmental threats from their disposal. The researchers recommend that policymakers “help to integrate the entire industry chain among automakers, battery producers, used-car dealers, and scrap companies in battery recycling systems to achieve a more sustainable society.”

This research was supported through the MIT Energy Initiative’s Mobility of the Future study.

This article appears in the Autumn 2020 issue of Energy Futures, the magazine of the MIT Energy Initiative. More

Not so long ago, watching a movie on a smartphone seemed impossible. Vivienne Sze was a graduate student at MIT at the time, in the mid 2000s, and she was drawn to the challenge of compressing video to keep image quality high without draining the phone’s battery. The solution she hit upon called for co-designing energy-efficient circuits with energy-efficient algorithms.

Sze would go on to be part of the team that won an Engineering Emmy Award for developing the video compression standards still in use today. Now an associate professor in MIT’s Department of Electrical Engineering and Computer Science, Sze has set her sights on a new milestone: bringing artificial intelligence applications to smartphones and tiny robots.

Her research focuses on designing more-efficient deep neural networks to process video, and more-efficient hardware to run those applications. She recently co-published a book on the topic, and will teach a professional education course on how to design efficient deep learning systems in June.

On April 29, Sze will join Assistant Professor Song Han for an MIT Quest AI Roundtable on the co-design of efficient hardware and software moderated by Aude Oliva, director of MIT Quest Corporate and the MIT director of the MIT-IBM Watson AI Lab. Here, Sze discusses her recent work.

Q: Why do we need low-power AI now?

A: AI applications are moving to smartphones, tiny robots, and internet-connected appliances and other devices with limited power and processing capabilities. The challenge is that AI has high computing requirements. Analyzing sensor and camera data from a self-driving car consumes about 2,500 watts, but the computing budget of a smartphone is just about a single watt. Closing this gap requires rethinking the entire stack, a trend that will define the next decade of AI.

Q: What’s the big deal about running AI on a smartphone?

A: It means that the data processing no longer has to take place in the “cloud,” on racks of warehouse servers. Untethering compute from the cloud allows us to broaden AI’s reach. It gives people in developing countries with limited communication infrastructure access to AI. It also speeds up response time by reducing the lag caused by communicating with distant servers. This is crucial for interactive applications like autonomous navigation and augmented reality, which need to respond instantaneously to changing conditions. Processing data on the device can also protect medical and other sensitive records. Data can be processed right where they’re collected.

Q: What makes modern AI so inefficient?

A: The cornerstone of modern AI — deep neural networks — can require hundreds of millions to billions of calculations — orders of magnitude greater than compressing video on a smartphone. But it’s not just number crunching that makes deep networks energy-intensive — it’s the cost of shuffling data to and from memory to perform these computations. The farther the data have to travel, and the more data there are, the greater the bottleneck.

Q: How are you redesigning AI hardware for greater energy efficiency?

A: We focus on reducing data movement and the amount of data needed for computation. In some deep networks, the same data are used multiple times for different computations. We design specialized hardware to reuse data locally rather than send them off-chip. Storing reused data on-chip makes the process extremely energy-efficient.  

We also optimize the order in which data are processed to maximize their reuse. That’s the key property of the Eyeriss chip that I co-designed with Joel Emer. In our followup work, Eyeriss v2, we made the chip flexible enough to reuse data across a wider range of deep networks. The Eyeriss chip also uses compression to reduce data movement, a common tactic among AI chips. The low-power Navion chip that I co-designed with Sertac Karaman for mapping and navigation applications in robotics uses two to three orders of magnitude less energy than a CPU, in part by using optimizations that reduce the amount of data processed and stored on-chip. 

Q: What changes have you made on the software side to boost efficiency?

A: The more that software aligns with hardware-related performance metrics like energy efficiency, the better we can do. Pruning, for example, is a popular way to remove weights from a deep network to reduce computation costs. But rather than remove weights based on their magnitude, our work on energy-aware pruning suggests you can remove the more energy-intensive weights to improve overall energy consumption. Another method we’ve developed, NetAdapt, automates the process of adapting and optimizing a deep network for a smartphone or other hardware platforms. Our recent followup work, NetAdaptv2, accelerates the optimization process to further boost efficiency.

Q: What low-power AI applications are you working on?

A: I’m exploring autonomous navigation for low-energy robots with Sertac Karaman. I’m also working with Thomas Heldt to develop a low-cost and potentially more effective way of diagnosing and monitoring people with neurodegenerative disorders like Alzheimer’s and Parkinson’s by tracking their eye movements. Eye-movement properties like reaction time could potentially serve as biomarkers for brain function. In the past, eye-movement tracking took place in clinics because of the expensive equipment required. We’ve shown that an ordinary smartphone camera can take measurements from a patient’s home, making data collection easier and less costly. This could help to monitor disease progression and track improvements in clinical drug trials.

Q: Where is low-power AI headed next?

A: Reducing AI’s energy requirements will extend AI to a wider range of embedded devices, extending its reach into tiny robots, smart homes, and medical devices. A key challenge is that efficiency often requires a tradeoff in performance. For wide adoption, it will be important to dig deeper into these different applications to establish the right balance between efficiency and accuracy. More