Can an aging corporation’s adventures in fundamental physics research open a new era of unimaginably powerful computers?
In 2012, physicists in the Netherlands announced a discovery in particle physics that started chatter about a Nobel Prize. Inside a tiny rod of semiconductor crystal chilled cooler than outer space, they had caught the first glimpse of a strange particle called the Majorana fermion, finally confirming a prediction made in 1937. It was an advance seemingly unrelated to the challenges of selling office productivity software or competing with Amazon in cloud computing, but Craig Mundie, then heading Microsoft’s technology and research strategy, was delighted. The abstruse discovery—partly underwritten by Microsoft—was crucial to a project at the company aimed at making it possible to build immensely powerful computers that crunch data using quantum physics. “It was a pivotal moment,” says Mundie. “This research was guiding us toward a way of realizing one of these systems.”
Microsoft
is now almost a decade into that project and has just begun to talk
publicly about it. If it succeeds, the world could change dramatically.
Since the physicist Richard Feynman first suggested the idea of a
quantum computer in 1982, theorists have proved that such a machine
could solve problems that would take the fastest conventional computers
hundreds of millions of years or longer. Quantum computers might, for
example, give researchers better tools to design novel medicines or
super-efficient solar cells. They could revolutionize artificial
intelligence.
Progress toward that computational nirvana has been
slow because no one has been able to make a reliable enough version of
the basic building block of a quantum computer: a quantum bit, or qubit,
which uses quantum effects to encode data. Academic and government
researchers and corporate labs at IBM and Hewlett-Packard have all built
them. Small numbers have been wired together, and the resulting devices
are improving. But no one can control the physics well enough for these
qubits to serve as the basis of a practical general-purpose computer.
Microsoft
has yet to even build a qubit. But in the kind of paradox that can be
expected in the realm of quantum physics, it may also be closer than
anyone else to making quantum computers practical. The company is
developing a new kind of qubit, known as a topological qubit, based
largely on that 2012 discovery in the Netherlands. There’s good reason
to believe this design will be immune from the flakiness plaguing
existing qubits. It will be better suited to mass production, too. “What
we’re doing is analogous to setting out to make the first transistor,”
says Peter Lee, Microsoft’s head of research. His company is also
working on how the circuits of a computer made with topological qubits
might be designed and controlled. And Microsoft researchers working on
algorithms for quantum computers have shown that a machine made up of
only hundreds of qubits could run chemistry simulations beyond the
capacity of any existing supercomputer.
In the next year or so,
physics labs supported by Microsoft will begin testing crucial pieces of
its qubit design, following a blueprint developed by an outdoorsy math
genius. If those tests work out, a corporation widely thought to be
stuck in computing’s past may unlock its future.
Stranger still: a physicist at the fabled but faded Bell Labs might get there first.
Tied Up in Knots
In
a sunny room 100 yards from the Pacific Ocean, Michael Freedman, the
instigator and technical mastermind of Microsoft’s project, admits to
feeling inferior. “When you start thinking about quantum computing, you
realize that you yourself are some kind of clunky chemical analog
computer,” he says. Freedman, who is 63, is director of Station Q, the
Microsoft research group that leads the effort to create a topological
qubit, working from a dozen or so offices on the campus of the
University of California, Santa Barbara. Fit and tanned, he has dust on
his shoes from walking down a beach path to lunch.
If his mind is a
clunky chemical computer, it is an extraordinary one. A mathematical
prodigy who entered UC Berkeley at the age of 16 and grad school two
years later, Freedman was 30 when he solved a version of one of the
longest-standing problems in mathematics, the Poincaré conjecture.
He worked it out without writing anything down, visualizing the
distortion of four-dimensional shapes in his head. “I had seen my way
through the argument,” Freedman recalls. When he translated that inner
vision into a 95-page proof, it earned the Fields Medal, the highest
honor in mathematics.
That cemented Freedman’s standing as a leading light in topology, the
discipline concerned with properties of shapes that don’t change when
those shapes are distorted. (An old joke has it that topologists can’t
distinguish a coffee cup from a doughnut—both are surfaces punctured by a
single hole.) But he was drawn into physics in 1988 after a colleague
discovered a connection between some of the math describing the topology
of knots and a theory explaining certain quantum phenomena. “It was a
beautiful thing,” says Freedman. He immediately saw that this connection
could allow a machine governed by that same quantum physics to solve
problems too hard for conventional computers. Ignorant that the concept
of quantum computing already existed, he had independently reinvented
it.
Freedman kept working on that idea, and in 1997 he joined
Microsoft’s research group on theoretical math. Soon after, he teamed up
with a Russian theoretical physicist, Alexei Kitaev, who had proved
that a “topological qubit” formed by the same physics could be much more
reliable than qubits that other groups were building. Freedman
eventually began to feel he was onto something that deserved attention
beyond his rarefied world of deep math and physics. In 2004, he showed
up at Craig Mundie’s office and announced that he saw a way to build a
qubit dependable enough to scale up. “I ended up sort of making a
pitch,” says Freedman. “It looked like if you wanted to start to build
the technology, you could.”
Mundie bought it. Though Microsoft
hadn’t been trying to develop quantum computers, he knew about their
remarkable potential and the slow progress that had been made toward
building them. “I was immediately fascinated by the idea that maybe
there was a completely different approach,” he says. “Such a form of
computing would probably turn out to be the basis of a transformation
akin to what classical computing has done for the planet in the last 60
years.” He set up an effort to create the topological qubit, with a
slightly nervous Freedman at the helm. “Never in my life had I even
built a transistor radio,” Freedman says.
Distant Dream
In
some ways, a quantum computer wouldn’t be so different from a
conventional one. Both deal in bits of data represented in binary form.
And both types of machine are made up of basic units that represent bits
by flipping between different states like a switch. In a conventional
computer, every tiny transistor on a chip can be flipped either off to
signify a 0 or on for a 1. But because of the quirky
rules of quantum physics, which govern the behavior of matter and energy
at extremely tiny scales, qubits can perform tricks that make them
exceedingly powerful. A qubit can enter a quantum state known as
superposition, which effectively represents 0 and 1 at
the same time. Once in a superposition state, qubits can become linked,
or “entangled,” in a way that means any operation affecting one
instantly changes the fate of another. Because of superposition and
entanglement, a single operation in a quantum computer can execute parts
of a calculation that would take many, many more operations for an
equivalent number of ordinary bits. A quantum computer can essentially
explore a huge number of possible computational pathways in parallel.
For some types of problems, a quantum computer’s advantage over a
conventional one grows exponentially with the amount of data to be
crunched. “Their power is still an amazement to me,” says Raymond Laflamme,
executive director of the Institute for Quantum Computing at the
University of Waterloo, in Ontario. “They change the foundation of
computer science and what we mean by what is computable.”
In the next year or so, physics labs supported by Microsoft will begin testing its qubit design.
But
pure quantum states are very fragile and can be observed and controlled
only in carefully contrived circumstances. For a superposition to be
stable, the qubit must be shielded from seemingly trivial noise such as
random bumping from subatomic particles or faint electrical fields from
nearby electronics. The two best current qubit technologies represent
bits in the magnetic properties of individual charged atoms trapped in
magnetic fields or as the tiny current inside circuits of
superconducting metal. They can preserve superpositions for no longer
than fractions of a second before they collapse in a process known as
decoherence. The largest number of qubits that have been operated
together is just seven.
Since 2009, Google has been testing a
machine marketed by the startup D-Wave Systems as the world’s first
commercial quantum computer, and in 2013 it bought a version of the
machine that has 512 qubits. But those qubits are hard-wired into a
circuit for a particular algorithm, limiting the range of problems they
can work on. If successful, this approach would create the
quantum-computing equivalent of a pair of pliers—a useful tool suited to
only some tasks. The conventional approach being pursued by Microsoft
offers a fully programmable computer—the equivalent of a full toolbox.
And besides, independent researchers have been unable to confirm that
D-Wave’s machine truly functions as a quantum computer. Google recently
started its own hardware lab to try to create a version of the technology that delivers.
The
search for ways to fight decoherence and the errors it introduces into
calculations has come to dominate the field of quantum computing. For a
qubit to truly be scalable, it would probably need to accidentally
decohere only around once in a million operations, says Chris Monroe, a professor at the University of Maryland and co-leader of a quantum computing project
funded by the Department of Defense and the Intelligence Advanced
Research Projects Activity. Today the best qubits typically decohere
thousands of times that often.
Microsoft’s Station Q might have a
better approach. The quantum states that lured Freedman into
physics—which occur when electrons are trapped in a plane inside certain
materials—should provide the stability that a qubit builder craves,
because they are naturally deaf to much of the noise that destabilizes
conventional qubits. Inside these materials, electrons take on strange
properties at temperatures close to absolute zero, forming what are
known as electron liquids. The collective quantum properties of the
electron liquids can be used to signify a bit. The elegance of the
design, along with grants of cash, equipment, and computing time, has
lured some of the world’s leading physics researchers to collaborate
with Microsoft. (The company won’t say what fraction of its $11 billion
annual R&D spending goes to the project.)
The catch is that
the physics remains unproven. To use the quantum properties of electron
liquids as bits, researchers would have to manipulate certain particles
inside them, known as non-Abelian anyons, so that they loop around one
another. And while physicists expect that non–Abelian anyons exist, none
have been conclusively detected.
Majorana particles, the kind of
non-Abelian anyons that Station Q and its collaborators seek, are
particularly elusive. First predicted by the reclusive Italian physicist
Ettore Majorana in 1937, not long before he mysteriously disappeared,
they have captivated physicists for decades because they have the unique
property of being their own antiparticles, so if two ever meet, they
annihilate each other in a flash of energy.
No one had reported
credible evidence that they existed until 2012, when Leo Kouwenhoven at
Delft University of Technology in the Netherlands, who had gotten
funding and guidance from Microsoft, announced that he had found them
inside nanowires made from the semiconductor indium antimonide. He had
coaxed the right kind of electron liquid into existence by connecting
the nanowire to a chunk of superconducting electrode at one end and an
ordinary one at the other. It offered the strongest support yet for
Microsoft’s design. “The finding has given us tremendous confidence that
we’re really onto something,” says Microsoft’s Lee. Kouwenhoven’s group
and other labs are now trying to refine the results of the experiment
and show that the particles can be manipulated. To speed progress and
set the stage for possible mass production, Microsoft has begun working
with industrial companies to secure supplies of semiconductor nanowires
and the superconducting electronics that would be needed to control a
topological qubit.
For all that, Microsoft doesn’t yet have its
qubit. A way must be found to move Majorana particles around one another
in the operation needed to write the equivalent of 0s and 1s.
Materials scientists at the Niels Bohr Institute in Copenhagen recently
found a way to build nanowires with side branches, which could allow
one particle to duck to the side while another passes. Charlie Marcus, a
researcher there who has worked with Microsoft since its first design,
is now preparing to build a working system with the new wires. “I would
say that is going to keep us busy for the next year,” he says.
Success
would validate Microsoft’s qubit design and put an end to recent
suggestions that Kouwenhoven may not have detected the Majorana particle
in 2012 after all. But John Preskill, a professor of theoretical
physics at Caltech, says the topological qubit remains nothing more than
a nice theory. “I’m very fond of the idea, but after some years of
serious effort there’s still no firm evidence,” he says.
Competitive Physics
At Bell Labs in New
Jersey, Bob Willett says he has seen the evidence. He peers over his
glasses at a dull black crystal rectangle the size of a fingertip. It
has hand-soldered wires around its edges and fine zigzags of aluminum on
its surface. And in the middle of the chip, in an area less than a
micrometer across, Willett reports detecting non-Abelian anyons. If he
is right, Willett is farther along than anyone who is working with
Microsoft. And in his series of small, careworn labs, he is now
preparing to build what—if it works—will be the world’s first
topological qubit. “We’re making the transition from the science to the
technology now,” he says. His effort has historical echoes. Down the
corridor from his labs is a glass display case with the first transistor
inside, made on this site in 1947.
Willett’s device is a version
of a design that Microsoft has mostly given up on. By the time the
company’s project began, Freedman and his collaborators had determined
that it should be possible to build a topological qubit using crystals
of ultrapure gallium arsenide that trap electrons. But in four years of
experiments, the physics labs supported by Microsoft didn’t find
conclusive evidence of non-Abelian anyons. Willett had worked on similar
physics for years, and after reading a paper of Freedman’s on the
design, he decided to have a go himself. In a series of papers published
between 2009 and 2013, he reported finding those crucial particles in
his own crystal-based devices. When one crystal is cooled with liquid
helium to less than 1 Kelvin (−272.15 °C) and subjected to a magnetic
field, an electron liquid forms at its center. Willett uses electrodes
to stream the particles around its edge; if they are non-Abelian anyons
looping around their counterparts in the center, they should change the
topological state of the electron liquid as a whole. He has published
results from several different experiments in which he saw telltale
wobbles, which theorists had predicted, in the current of those flowing
particles. He’s now moved on to building a qubit design. It is not much
more complex than his first experiment: just two of the same circuits
placed back to back on the same crystal, with extra electrodes that link
electron liquids and can encode and read out quantum states that
represent the equivalent of 0s and 1s.
Willett
hopes that device will squelch skepticism about his results, which no
one else has been able to replicate. Microsoft’s collaborator Charlie
Marcus says Willett “saw signals that we didn’t see.” Willett counters
that Marcus and others have made their devices too large and used
crystals with important differences in their properties. He says he
recently confirmed that by testing some devices made to the
specifications used by other researchers. “Having worked with the
materials they’re working with, I can see why they stopped doing it,
because it is a pain in the ass,” he says.
Bell Labs, now owned by the French telecommunications company
Alcatel-Lucent, is smaller and poorer than it was back when AT&T,
unchallenged as the American telephone monopoly, let many researchers do
pretty much anything they desired. Some of Willett’s rooms overlook the
dusty, scarred ground left when an entire wing of the lab was
demolished this year. But with fewer people around than the labs had
long ago, it’s easier to get access to the equipment he needs, he says.
And Alcatel has begun to invest more in his project. Willett used to
work with just three other physicists, but recently he began
collaborating with mathematicians and optics experts too. Bell Labs
management has been asking about the kinds of problems that might be
solved with a small number of qubits. “It’s expanding into a relatively
big effort,” he says.
Willett sees himself as an academic
colleague of the Microsoft researchers rather than a corporate
competitor, and he still gets invited to Freedman’s twice-yearly
symposiums that bring Microsoft collaborators and other leading
physicists to Santa Barbara. But Microsoft management has been more
evident at recent meetings, Willett says, and he has sometimes felt that
his being from another corporation made things awkward.
It would
be more than just awkward if Willett beat Microsoft to proving that the
idea it has championed can work. For Microsoft to open up a practical
route to quantum computing would be surprising. For the withered Bell
Labs, owned by a company not even in the computing business, it would be
astounding.
Quantum Code
On Microsoft’s
leafy campus in Redmond, Washington, thousands of software engineers
toil to fix bugs and add features to Windows and Microsoft Office.
Tourists pose in the company museum for photos with a life-size cutout
of a 1978 Bill Gates and his first employees. In the main research
building, Krysta Svore leads a dozen people working on software for
computers that may never exist. The team is figuring out what the first
generation of quantum computers could do for us.
The group was
established because although quantum computers would be powerful, they
cannot solve every problem. And only a handful of quantum algorithms
have been developed in enough detail to suggest that they could be
practical on real hardware. “Quantum computing is possibly very
disruptive, but we need to understand where the power is,” Svore says.
“We believe that there’s a chance to do something that could be the foundation of a whole new economy.”
No
quantum computer is ever going to fit into your pocket, because of the
way qubits need to be supercooled (unless, of course, someone uses a
quantum computer to design a better qubit). Rather, they would be used
like data centers or supercomputers to power services over the Internet,
or to solve problems that allow other technologies to be improved. One
promising idea is to use quantum computers for superpowered chemistry
simulations that could accelerate progress on major problems in areas
such as health or energy. A quantum computer could simulate reality so
precisely that it could replace years of plodding lab work, says Svore.
Today roughly a third of U.S. supercomputer time is dedicated to
simulations for chemistry or materials science, according to the
Department of Energy. Svore’s group has developed an algorithm that
would let even a first-generation quantum computer tackle much more
complex problems, such as virtually testing a catalyst for removing
carbon dioxide from the atmosphere, in just hours or minutes. “It’s a
potential killer application of quantum computers,” she says.
But
it’s possible to envision countless other killer applications. Svore’s
group has produced some of the first evidence that quantum computers can
be used for machine learning, a technology increasingly central to
Microsoft and its rivals. Recent advances in image and speech
recognition have triggered a frenzy of new research in artificial
intelligence. But they rely on clusters of thousands of computers
working together, and the results still lag far behind human
capabilities. Quantum computers might overcome the technology’s
limitations.
Work like that helps explain how the first company to
build a quantum computer might gain an advantage virtually
unprecedented in the history of technology. “We believe that there’s a
chance to do something that could be the foundation of a whole new
economy,” says Microsoft’s Peter Lee. As you would expect, he and all
the others working on quantum hardware say they are optimistic. But with
so much still to do, the prize feels as distant as ever. It’s as if
qubit technology is in a superposition between changing the world and
decohering into nothing more than a series of obscure research papers.
That’s the kind of imponderable that people working on quantum
technology have to handle every day. But with a payoff so big, who can
blame them for taking a whack at it?
No comments:
Post a Comment