This afternoon, MIT’s Political Science distinguished speakers series hosts Regina Dugan and Kaigham Gabriel, director and deputy director of DARPA, the US defense advanced research project agency, who are here to speak about advanced manufacturing in America. The title for their talk is “Just Make It”, a response Dugan offers to people who ask her to predict the future. “Visionaries aren’t oracles – they are builders.”
She shows a five minute video of nerd porn, a montage of dismissive predictions about technologies (like Lord Kelvin’s statement about the impossibility of heavier than air flight, followed by footage of the Wright Brothers, and then from Top Gun. The video ends with observations about the time to 50 million users for different technologies is rapidly shrinking, pointing to Facebook’s sprint to 100 million users, and offers images of protesters holding banners celebrating the internet. “Still think social media is a fad?” the video asks. The video ends with a challenge for the engineers in the room – “just make it”.
Dugan tells us that the decline in America’s ability to build things is a national challenge, if not a crisis. Americans consume an increasing percentage of goods made overseas, and are less likely to be employed making things. Perhaps this reflects on productivity increases, or on currency manipulations, but it has implications, she warns, for national defense. Adam Smith warned that if an industry was critical to defense, it is not always prudent to rely on neighbors for supply.
There have been many years of debate around the inefficiency of America’s design and building of defense systems, Dugan tells us. One extrapolation of increase in airline design cost – sometimes referred to as “Augustine’s Laws” – suggests that by 2054, a single military aircraft will cost as much as the entire military budget at that time. Obviously, it’s dangerous to extrapolate linearly from current data… but if you do, the cost of military systems is growing much more rapidly than defense budgets. “Quite obviously, this is not sustainable”.
When we design aircraft, she tells us, we’re often designing ten years out. That means we’re trying to understand the threat environment ten years out. That’s risky. “Lack of adaptability is a vulnerability.”
What’s worse is that it’s really expensive. She shows a graph of production costs for the F-22 fighter. The price per unit keeps increasing, and the volume required keeps dropping. This might be because we need to amortize design costs over fewer units. Or it might be because the costs get so high, we simply can’t afford as many units as we wanted. This isn’t just true of the F-22 – it’s true for the Marine EFV project and the Comanche helicopter as well.
This difficulty in building complex systems has implications for defense and for the economy as a whole, she tells us.
“To innovate, we must make. To protect, we must produce.” DARPA is not a policy organization, she tells us, but pushing from “a buy to make strategy” is of strategic importance to the US Department of Defense.
There’s $200 million a year being invested in innovation, looking for ways to change the calculus of cost increase. Can we turn a long problem like vaccine design into one we can solve in weeks? Could we permit the participation of tens of thousands of designers into a process and harness their ideas? She suggests that the future of innovation is around increased speed of production and number and diversity of designs. The rise of electronic design aides revolutionized the semiconductor industry – could this shift in speed and diversity bring a similar paradigm shift?
Dugan tells us that the systems we have to manage complexity are inherited from 1969-era systems engineering. We take complex systems and split them along functional lines – power system, control system, thermal system – then try to put them back together. What happens is that we experience emergent behaviors that weren’t predictable. As a result, we end up with a design, build and test system that we iterate through, trying to solve those emergent problems.
This isn’t the only way to design complex systems. She shows a graph that measures time to design, integrate and test, versus a measure of product complexity, which includes part count and lines of source code. There’s a linear increase in time to build to complexity for aerospace defense systems. Another piece of the graph shows a flat design and test time cycle with increasing complexity – that’s the semiconductor industry. And a third industry – the best in class automotive manufacturers – show a decrease in time with an increase in complexity! How are they pulling this off?
Gabriel tags in here, to explain how the semiconductor industry achieved gains in complexity without extending the timeframe necessary to design and test their products. The key factor was a decision to control for time. “If we aren’t out there with new chips in 18-24 months, we’ll miss the next generation of PCs.” So the principles of VLSI design were optimized around producing new product on a timecycle as tight as that for less complex integrated circuits.
Two major design innovations characterize the VLSI shift, Gabriel tells us. First, it’s critical to decouple design and fabrication, a shift that was comparatively easy for circuit designers to accept. The second was initially heresy: you needed to stop optimizing each transistor, and sacrifice component performance for ease of system design and reliability.
We’ve seen a similar move in computer programming, a shift away from assembler, which produces very efficient code that’s hard to test, to higher level programming languages. Those languages abstract operations, which leads to a decrease in performance efficiency, but since we’re no longer as limited by how many operations a computer can perform, the design speed benefits outweigh the performance compromises. He hints that we may be seeing some similar shifts in biological sciences as well.
How does this work in terms of DARPA projects? Dugan retrieves the mic to speak about the Adaptive Vehicle Make program, designed to build a new infantry vehicle in two years instead of ten. A first step is developing a language to describe and design mechano-electric systems so they can integrate more smoothly. The vehicle, she tell us, will be flexibly manufactured through a “bitstream-configurable foundry-like manufacturing capability for defense systems” capable of “mass production in quantities of one”.
With facilities that can accept a design and custom-forge parts, she believes we can move to an increasingly democratized design system, which enables the participation of many more people to design and submit systems to foundry-like fabrication facilities. We’ll design vehicles “using the most modern techniques of crowd infrastructure and open source development,” in a program called VehicleForge.mil. (While a valid URL, there’s no webserver at that address. Just wanted to save you a Google search or two.)
Critics tell her this approach won’t work. But Dassault recently designed the Falcon 7x aircraft using “digital master models, by tail number, for aircraft” – i.e., building extremely complex individual models for each aircraft they build. The models only do geometric interference (i.e., they test whether the parts fit together), but they’ve halved the time needed to produce a new plane. Critics claim that the analogy between integrated circuits and military vehicles is an inept one. But in terms of part count, ICs are much more complex than vehicles. What’s complex is the diversity of components used in the combat vehicle.
A new experiment, conducted in cooperation with Local Motors, a small-scale vehicle fabrication company (see my notes on the founder’s Pop!Tech talk in 2009) invites designers to compete to design a combat support vehicle, the XC2V. $10,000 in prizes were offered, and instead of getting the 3 designs they get in an invitation-only design scenario, they received 159, 100 of which the judges deemed “high calibre”. It wasn’t a clean sheet of paper design – the chassis and drivetrain were designed by Local Motors – but it was effective at expanding the idea pool, and led to a functioning design within four weeks.
The power of the crowd may be even greater in a field like protein folding, where humans are still able to solve some problems better than algorithms. Foldit is the brainchild of a biochemist, a computer scientist and a gamer, who decided to turn protein folding into a game, building “a Tetris-like environment for folding”. 240,000 people have signed up to play, but what’s really cool is “the emergence of 5 sigma savants for protein folding, some of whom have very little biochemistry training.” Recently, Foldit solved a key protein – a retroviral protease SIV for the rhesus monkeys – which had been unsolved for 15 years. The community folded it in 10 days. Projects like this, she tells us, make her a believer that bringing many diverse minds to a problem and increasing the pace of building will increase the speed and diversity of innovation.
Gabriel offers three other examples where massive innovations are possible through new methods.
Optics are the dominant cost in many imaging and sensor systems. It turns out that making light do something different – bending, focusing, diffusing – requires materials and systems that are heavy, complex and expensive. M-GRIN – manufacturable gradient index optics – moves beyond lenses that are made out of a single material with a single index of refraction. Instead, they use a stack of multiple layers and films, combined via heat and pressure, to make lenses that are smaller and lighter. A test around a shortwave infrared lens produced a device that was 3.5x smaller and 7.5x lighter. That’s a breakthrough… but the real innovation is creation of a set of design rules that let you go from an application to a recipe for combining materials into the lens you need.
In telling us about maskless nanolithography, Gabriel tells us “Moore’s law is dead in circuit design, though the corpse doesn’t know it yet.” The culprit is heat – we can make tighter and smaller circuits, but they’re getting very difficult to cool. As critical is cost. Working at ultra-small line width is prohibitively expensive. It’s hard to spend tens of millions on a set of 45 nanometer masks to create a few hundred chips for a defense system, when building those masks costs tens of millions of dollars.
We know how to do lithography without masks, but it’s traditionally been very slow. So now designers have built a system that creates and bends an electron beam, then splits it into millions of beamlets, controlled by a “dynamic pattern generator”. Program that pattern generator, and it allows millions of writing operations to happen at the same time, leading to a current working speed of 10-15 wafers per hour, the minimum required to produce custom ICs for military applications.
His third example is the accelerated manufacturing of pharmaceuticals, a strategy he tells us was Plan B in 2009-2010 if the H1N1 flu virus had resurfaced. It’s very hard to produce vaccines quickly – egg-based strategies require a piece of virus and many thousands of chicken eggs. These methods work, but can require 6-9 months to build up a stockpile. A new method uses tobacco plants to produce custom proteins, working from strands of DNA in the virus. Envision a football-field sized building filled with lights and trays of tobacco plants. A facility like that can now produce a million doses a month of a novel vaccine. In scaling up capacity to 100 million doses per month, the key problem turned out to be lighting – it was impossible to light everything without switching to LED bulbs. Once they made the switch, they had a new opportunity – tuning the spectrum to optimize production. Using an experiment of “high school science complexity”, they grew plants under different lighting conditions for a few weeks, and determined a mix of blue and red frequencies that doubles protein production.
Gabriel ends with a slide quoting MIT scientist Tom Knight:
“The 19th century was about energy.
The 20th century was about information.
The 21st century is about matter.”
If we embrace this challenge, Gabriel tells us, we will be able to make things at the cost we used to produce and stockpile them in bulk, and this change will change how we innovate.
Above this line are my notes, below, my reaction:
I thought the DARPA folks gave an impressive talk, inasmuch as they got me thinking about a problem I’d not considered – the insane cost and time frame of producing military equipment. But for a talk sponsored by the political science department, it seemed woefully lacking of discussions of politics or markets. If I were trying to explain the difference in production processes between military vehicles, consumer automobiles and integrated circuits, I suspect I might look at the power of markets. IC manufacturers needed to build chips quickly because customers wanted to buy newer, faster chips… and would buy other chips if the manufacturer wasn’t fast enough. Ditto for automobile companies.
The defense industry is different. It’s very hard to terminate a weapons system, even if it’s massively over time and over budget. The competition happens well before a product is built. Discovering that the F-22 production isn’t going well doesn’t create a market opportunity for another company to produce a better product faster – the company producing the F-22 is going to get paid, even if they take an absurd time to produce the product.
I admire the approach Dugan and Gabriel are putting forward, and certainly appreciate that it plays well to a room full of engineers. But I was very surprised not to hear questions (and I only caught the first five or six) about whether the DoD purchasing process can be reformed so long as military budgets are sacrosanct. We’re currently facing mandatory budget cuts with the failure of the budget supercommmittee, and conventional political wisdom suggests that the social service cuts will go through, while the defense ones will not. How do you encourage companies to innovate when they’re currently amply rewarded for dragging design and production out over decades? How do you innovate without market pressures?
My homogeneously left-wing family was talking politics over the Thanksgiving dinner table and realized the solution to America’s current social problems was to simply adopt the Egyptian political system – let the military run everything. The right doesn’t like cutting military budgets, but is okay when the military provides state-sponsored healthcare and subsidizes education. All we need to do is ensure all Americans are employed by the US military and we can build a thriving, successful welfare state. The same absurdity behind that suggestion is what makes DARPA’s ideas so hard to implement – if there’s no pressure to cut military budgets, anything is possible… except real innovation around cost and efficiency.