Energy, Security, and Climate

On Monday, the Obama administration gave Shell conditional permission to move forward with Arctic oil drilling. The New York Times captures a common sentiment well in identifying this as a “tricky intersection of Obama’s energy and climate legacies”. The reality, though, is that this intersection isn’t nearly a fraught as many assume: decisions about offshore drilling in Alaska are indeed difficult, given the local economic and environmental stakes involved, but climate isn’t a central factor. I’m ambivalent when it comes to federal decisions on offshore Arctic drilling. The Arctic is a special place. I saw that first hand when I visited with the Coast Guard in 2008 – a trip on which I also learned how challenging oil spill response there can be. (I also learned that a buoy tender isn’t the ideal place to spend your first night ever at sea.) Opposing offshore Arctic oil development is a reasonable position. At the same time, with the right precautions, spill risks can be substantially reduced, though inevitably not eliminated. And there’s a federalism issue (perhaps not in the legal sense but in a more basic one): it’s easy to be strident in taking positions from Washington, DC, but this is a much more intimate economic and environmental issue for Alaskans – so presumably their preferences should have some special say. Navigating these tradeoffs is difficult. But throwing climate change into the mix as a central consideration lacks empirical foundation. (Perhaps that’s why that Times article doesn’t follow through on its headline’s promise.) Yes, at a global level, more oil production means more oil consumption, and hence greater carbon dioxide emissions and worse climate change. But more oil production in one place generally means less oil production elsewhere – that’s how markets and prices work – which substantially blunts the effect. Bill McKibben drills home the conventional wisdom in a Times op-ed, blaming Obama for “climate denial” by claiming that “you can’t deal with climate on the demand side alone”, backing that up by citing a study that was unable to identify any “climate-friendly scenario in which any oil or gas could be drilled in the Arctic”. True! Also true: that claim was based on looking at a whopping two scenarios. (From the original: “none [of the oil or gas] is produced in any [Arctic] region in either of the 2C scenarios before 2050”.) And, most important, the study never asked what would happen to emissions if the Arctic oil were put off limits. Had it done so, it would have found more oil production elsewhere, and minimal net emissions impact. What the study really found – and what is entirely reasonable – is that if the world gets serious about reducing emissions, oil prices will fall, and companies won’t want to develop most Arctic oil anyhow. That points to demand-side policy, denigrated by many who are painting the Alaska decision as climate apostasy, as critical. There is one theoretical exception. The United States, Saudi Arabia, Russia, Iran, and a bunch of other oil producers could team up to jointly restrict oil production. That prospect, of course, makes U.S.-China-India-Europe cooperation to reduce emissions through demand-side policy look like a cakewalk by comparison. Navigating the local economic and environmental tradeoffs involved in Arctic oil development is difficult enough without turning every decision into a climate litmus test. And getting serious on climate change is plenty tough without pretending that playing fossil fuel whack-a-mole whenever possible will be effective in reducing emissions. We’ll have better policy, and better outcomes, if we don’t make every difficult energy and environment decision about climate change too.

Over the weekend, Moore’s Law—the prediction that the number of transistors (building blocks) on an integrated circuit (computer chip or microchip) would double every two years—turned fifty years old. It so happens that the silicon solar panel, the dominant variety in the market today, is about the same age—roughly fifty-two years old. And over the last half-century, while the computing power of an identically sized microchip increased by a factor of over a billion, the power output of an identically sized silicon solar panel more or less doubled.[1] The contrast between Moore’s Law for microchips and the plodding progress of clean technology is bittersweet for me. Growing up, I would wait impatiently for my father to bring home a new computer, powered by a faster, next-generation processor—his office was across from Gordon Moore’s at Intel. When he founded a solar panel start-up, he brought the limitless optimism of Moore’s Law with him, but like myriad other cleantech start-ups, his company struggled to stay afloat given the surging tide of cheap, mediocre silicon solar panels from China.[2] My own doctoral research focused on exciting alternatives to silicon solar panels, but those alternatives face a daunting barrier to entry from large silicon firms. So while I celebrate the startling prescience with which Gordon Moore, in 1965, predicted the density of transistors fifty years hence and every year in between, I reject the notion that clean technologies like solar panels and batteries follow a Moore-esque decline in cost. Unfortunately, a chorus of voices in the mainstream media have echoed the claim that Moore’s Law is guiding the regular decline in clean technology costs as production increases, enabling a massive energy transition from fossil fuels. In an excellent 2011 piece, Michael Kanellos at Forbes gently corrected this claim, but he was still too charitable in conceding that clean energy advocates were “wrong in the particulars, but right in their outlook.” Rather, that outlook is far too complacent, satisfied with pedestrian cost declines and stagnating performance in lieu of disruptive technology advances, more in line with Moore’s Law. To date, there have been three crucial differences between Moore’s law for microchips and the historical cost declines of solar panels and batteries: 1. Moore’s Law is a consequence of fundamental physics. Clean technology cost declines are not. 2. Moore’s Law is a prediction about innovation as a function of time. Clean technology cost declines are a function of experience, or production. 3.Why this all matters. Moore’s Law provided a basis to expect dramatic performance improvements that shrank mainframes to mobile phones. Clean technology cost declines do not imply a similar revolution in energy. Difference #1: Moore’s Law is a consequence of fundamental physics. Clean technology cost declines are not. When Gordon Moore made his prediction in 1965 (original article in Electronics here), that the most economical number of transistors on a computer chip would double every two years, he based his reasoning on the physics of transistors. In hindsight, Moore’s physical instinct was confirmed: “As the dimensions of a transistor shrank, the transistor became smaller, lighter, faster, consumed less power, and in most cases was more reliable…it has often been a life without tradeoffs.” In other words, Moore’s primary insight was that shrinking the transistor made it work better—as a happy corollary to this shrinkage, the cost per unit of computing power kept falling, because the cost of manufacturing the same chip area has remained roughly constant. On the contrary, clean technology cost declines have very little to do with the physics of the actual devices being built. For example, falling costs of silicon solar panels have largely been driven by lower input material costs from scale, lower labor costs through manufacturing automation, and lower waste driven by efficient processing. All of these cost reductions follow naturally from manufacturing scale and vertical integration, rather than performance improvements. Now Tesla hopes to achieve similar cost reductions by building a Gigafactory—a massive facility in Nevada—to scale up production of lithium-ion batteries that will only perform incrementally better than the current generation. Difference #2: Moore made a prediction about innovation as a function of time. The advances of clean technology are a function of experience, or production. The physical principles that underpin favorable microchip shrinkage enabled Moore to predict that transistor density would double every two years, rather than predict progress based on the quantity of microchip production. Indeed, fulfilling Moore’s prophecy has required scientists and engineers to sit down at the drawing board and redesign each new generation of microchip down to the basic architecture of the transistor—Moore himself predicted that his law would require engineers to employ “cleverness.” Manufacturing advances, for example to enable printing ever more miniscule features on a chip, accompanied the design decisions to change the transistor architecture and chip design. Microchips certainly got cheaper as a result of manufacturing scale, but the increased production was never a sufficient condition to meet Moore’s Law—fundamental research and development (R&D) drove the advances. By contrast, clean energy devices basically look the same decade after decade. Instead of R&D advances, solar panels and batteries have come down in cost through the benefits of experience and scale, closely related to “learning by doing.” But this relationship between cost and production quantity—the “experience curve”—is a generic one cutting across many industries, from aerospace to chemicals. “Wright’s Law” and “Henderson’s Law” describe the inevitable cost reductions that accompany an industry’s increased experience through scale production of a good. These purported laws do not specify a numerical estimate for the experience curve’s slope—it happens to vary widely among industries. Although Moore’s Law is also, strictly speaking, not a law but an observation, it was extraordinary because it was a specific, quantitative prediction which the microchip industry proceeded to closely follow. Bloomberg New Energy Finance’s Michael Liebreich pointed out last week that the experience curves for solar and batteries look awfully similar, apparent from the similar slope estimates given on the slide. Perhaps this will enable a reliable prediction for the falling cost of batteries as they scale up. But whereas Moore could set a future-looking roadmap with real dates that the industry could target, setting a roadmap based on an experience curve requires perfect forecasting of future production, which is difficult, especially in a rapidly growing market.[3] **Why this all matters** Difference #3: Moore’s Law provided a basis to expect dramatic performance improvements that shrank mainframes to mobile phones. Clean technology cost declines do not imply a similar revolution in energy. To put this all together, Moore’s Law worked so well for microchips, because Moore had a set of physical reasons to believe in his prediction: that the density of transistors on a chip would inexorably increase as the years passed, because transistors just work better when they are smaller. And by making a specific, quantitative prediction, Moore’s Law became a self-fulfilling prophecy—industry leaders like Intel and Qualcomm published technology roadmaps that signaled their intention to match Moore’s Law so that financial markets and the rest of the computing ecosystem knew what to expect. As a result, electronic devices evolved in lockstep with the evolving, shrinking transistor, and today’s mobile phones have more memory and computing power than supercomputers in the 1980s. A different and generic phenomenon—the experience curve—characterizes solar and battery cost declines. If these cost declines were accompanied by performance enhancements, then an industry roadmap could spark synergistic innovation—new long-range electric vehicle designs might evolve alongside lighter batteries. But as costs decline while performance stagnates, it is not just a matter of time before an electric vehicle with a thousand mile range emerges; instead of leading energy innovation, clean technology firms merely peddle commodities. Today, the total cost of putting solar panels on a rooftop is over four times the cost of the panels themselves, so absent efficiency gains, the falling costs of solar panels, however predictable, only play a minor role in the final system cost. Moore’s Law is so special because the prediction it did make about the shrinking transistor resulted in a digital revolution that was completely unpredictable. Although Moore’s Law does not apply to the consistent cost declines of clean technologies, an appreciation for the dynamism that accompanied fifty years of regular progress in electronics could inspire advances on the energy front. Footnotes [1] This comparison is not apples to apples, because there is only so much of the sun’s energy that a solar panel can convert into electricity (its “efficiency”). Still, commercially dominant silicon panels are only around 15% efficient, only twice as efficient as Sharp’s inaugural 1963 solar module, whereas much higher efficiencies (>40%) have been demonstrated in labs and in the field. In other words, commercial solar panels are performing well below their potential today. To derive the computing power density improvement, I compared the 1963 IMB Gemini Digital Computer (~7,000 floating point operations per second (FLOPS)) with the 2013 GeForce GTX 780 r.2 chip (~4,000 GFLOPS) and adjusted for size. [2] Fittingly, my father went back to a world governed by Moore’s Law. He now leads research and development at SanDisk, and, confronted with the physical limits of shrinking transistors to just a few atoms wide, he is determined to keep pace with Moore’s Law by stacking transistors on top of each other. [3] One report notes that a Moore-esque time-based prediction and production-based experience curve predictions are actually equivalent under exponentially increasing production. Since clean technology production seems to be following exponential growth, in the near term it may be possible to forecast a roadmap in terms of dates rather than production quantities. However, the case of microchips is instructive—as soon as microchip production stopped growing exponentially and leveled off around the end of the 1990s, transistor shrinkage started to outpace the experience curve, demonstrating that Moore’s Law is fundamentally a function of time, not production.

The P5+1 and Iran have announced a framework for negotiating a final agreement to limit the Iranian nuclear program by the end of June. Here are five quick thoughts on the nuclear and sanctions elements: The nuclear limits – particularly those on the Iranian supply chain – are surprisingly strong and significant. The rough scale of Iranian enrichment activities and low-enriched uranium stockpiling that the United States could tolerate has long been pretty clear. (I laid out the basic logic for the sorts of limits that would allow the United States to respond effectively to Iranian breakout in a technical paper a few years ago, and many others have made similar arguments.) The announced framework tracks those understandings. What I’m struck by, though, is the extent of the monitoring provisions, particularly as they apply to the Iranian supply chain. U.S. policymakers have long feared that so long as Iran could conduct some legal nuclear commerce, it would be easier to hide illegal activities, making a secret parallel nuclear program more feasible. The framework includes some pretty strong steps to address this, including a “dedicated procurement channel” for the nuclear program. These may seem like footnotes compared to the rule for centrifuges and uranium stocks, but they’re central. The time it takes Iran to comply with the agreement will depend on its final details. Take one example: Iran is required “to reduce its current stockpile of about 10,000 kg of low-enriched uranium (LEU) to 300 kg of 3.67 percent LEU”. How will this be done? Will Iran ship the material out of country? Will it blend it down to LEU that’s enriched to less than 3.67 percent? Will it convert the LEU into fuel? (Which of these will the final deal allow?) The path it takes will determine how long compliance takes, which will affect the pace of sanctions removal. Similar questions surround many other provisions. It is unclear how sanctions relief will be phased in. If compliance occurs gradually over an extended period, sanctions relief will presumably be drawn out too. Iran faces a host of sanctions on oil sales, financial transactions, travel by senior officials, and other activities. Which sanctions will be pared back first? What milestones will they be connected to? Much of this presumably remains to be negotiated, but the details will be critical to determining the pace with which sanctions are removed – and, in particular, the speed at which full-scale Iranian oil exports come back online. Removing sanctions won’t necessarily lead to a rush back into Iran. The framework notes that sanction could be “snapped back” on Iranian noncompliance. Energy (and other) companies will presumably be slow to invest in Iran given the risk that they could easily find themselves faced with sanctions once again. Financial players may decide that the complexity and risks of dealing with Iran outweigh the limited commercial benefits. Oil traders, though, are more short term in nature, and will presumably reengage quickly. This experience is going to make U.S. policymakers even more sanctions-happy than before. An easy lesson of the Iran experience (presuming that the framework actually leads to a final deal) will be that sanctions, when combined with diplomacy, can yield meaningful results. Political scientists and policy analysts will doubtless debate this until eternity: one can’t say definitively what role sanctions played in bringing about the agreement, nor is there an objective way to know whether the framework deal is better than whatever else might have happened. The reality, though, is that many policymakers will take today’s news as a straightforward affirmation that sanctions work.