Real progress is being made using existing manufacturing approaches, but performance still has a long way to go.
For years, there has been an intense race among various nations to develop the world’s fastest supercomputers. The U.S. and Japan led the field until 2010, when China stunned the market and rolled out the world’s fastest supercomputer. And today, China continues to lead the field with a supercomputer capable of running at speeds of 33.86 petaflops per second.
While the supercomputer race continues to unfold, there is also a battle brewing in another high-end systems segment—quantum computing. In fact, a number of entities in Asia, Europe and the United States are accelerating their efforts to develop quantum computers, which supposedly are faster than today’s supercomputers.
Still in the R&D stage, quantum computers are making rapid progress. For example, the U.S. National Security Agency (NSA) is working on a quantum computer. Recently, startup D-Wave shipped the world’s first quantum computer to a joint venture between Google, NASA and a research group. In addition, IBM, R&D organizations and universities also are developing the technology.
In classical computing, the information is stored in bits, which can be either a “0” or “1”. Meanwhile, in quantum computing, information is stored in quantum bits, or qubits, which can exist as a “0” or “1” or a combination of both. Unlike a classical computer, which processes information in rapid succession, a quantum computer checks many different possibilities in parallel. The superposition state enables a quantum computer to perform millions of calculations at once. And increasing the number of qubits increases this parallelism exponentially.
But like today’s computers, the components used in quantum computers are silicon-based devices. Quantum devices are not leading-edge chips, but they are fabricated using today’s fabs and tools.
And thanks to a number of recent breakthroughs on the device side, quantum computers are moving one step closer towards their ultimate goal—these number-crunching systems could crack the world’s most complex algorithms within a reasonable time. This includes Shor’s algorithm, an integer factorization problem that can be utilized to break the widely used public-key cryptography scheme known as RSA.
“Quantum computers can solve that inverse problem faster,” said Thomas Theis, executive director at the Nanoelectronics Research Initiative (NRI), a program at the Semiconductor Research Corp. (SRC). “The U.S. doesn’t want another country having that capability. That’s driving a lot of funding (in quantum computing).”
Still, the question is whether quantum computing is a viable technology or mere hype. Many of the recent claims made by D-Wave are being disputed by IBM and other experts. And the development of a full-scale, “universal” quantum computer is still years, if not decades, away.
“There is still a long ways to go,” said Matthias Steffen, research staff member at IBM, who is leading the company’s experimental quantum computing group. “It’s hard to project out how far reaching the applications are in quantum computing. In addition, we also don’t know how many physical computation steps are required to do error correction. And we also don’t have a good handle on the clock speeds of a system.”
The soul of a quantum machine
Since the inception of the quantum computing field in the 1980s, a plethora of entities have been pursuing the technology, which is complex, confusing and often misunderstood. “Quantum computers will not necessarily outperform classical computers (in all tasks),” said Michelle Simmons, director of the Centre for Quantum Computation and Communication Technology (QCCT) at the University of New South Wales in Sydney, Australia.
Quantum computers excel at crunching specific algorithms. In one example, a salesman wants to travel to many cities in the shortest route. If the salesman traveled to 14 cities, that equates to 10 to the 11th power in terms of routes. A 1GHz classical computer could solve the problem in 100 seconds, Simmons said.
If the salesman traveled to 22 cities, that equates to 10 to 16th power in terms of routes. A 1GHz classical computer would not be able to solve the problem for 1,600 years, she said. “In quantum computing, it could be done much faster—in a practical time frame,” she said.
IBM’s Steffen framed the problem in a different light. “You can use a classical computer to solve problems. The number of resources required to solve the problem scales polynomially with the problem size,” he said. “Then, there is a category of problems that scales exponentially. Those are retractable. Even if you make the problem size just a little bit bigger, it basically becomes impossible to solve it. Remarkably, a quantum computer can solve certain problems that fall into that category and make them retractable.”
Still, in the community, there is a debate regarding the overall definition of a quantum computer. Last year, D-Wave shipped its 512-qubit quantum computer to the Quantum Artificial Intelligence Lab, a venture between NASA, Google and the Universities Space Research Association. Fabricated using a standard semiconductor process, D-Wave’s processors combine superconducting flux qubits and Josephson junctions.
The initial results from D-Wave’s system are mixed. “The system that D-Wave is pursing at the moment is not quantum-universal. For example, it is not able to do the factoring algorithm,” said IBM’s Steffen, referring to Shor’s algorithm.
What is a quantum computer?
In fact, IBM and others are developing true “universal” quantum computers. The term “universal” refers to a system that can harness the power of quantum computations, including Shor’s algorithm.
In theory, a 30-qubit, “universal” quantum computer could outperform today’s supercomputers. Today, IBM is developing a 100-qubit “universal” quantum computer. IBM, however, only has demonstrated a 3-qubit device. In fact, the world’s most advanced device is arguably a 14-qubit chip, meaning the technology is still in its infancy.
“If you want to have a system that can eventually do something useful, you probably need millions of qubits. There is a large gap between three and millions. And that needs to be closed,” Steffen said.
A 100-qubit system is also a potentially powerful computer. “The intrinsic device performance, as measured by the quantum life times of quantum states, is now sufficiently good enough that you can envision stringing as many as 10 qubits together. With 10 quibts, you can still get the right answer most of time,” he said. “We will see 10- to 100-qubit systems over the next several years. The real big thing here is the ability to demonstrate a logical qubit. A logical qubit is one that performs error correction. But no one has been able to show this in the lab. That would be the Holy Grail that would lay the foundation towards building larger systems.”
Obviously, the industry requires a number of breakthroughs. “The question is how do you put 10 or 100 qubits on a chip?” he said. “How do you control them? How do you make sure that the device performance does not deteriorate? And how do you write the software for this?”
Over the years, however, the industry has demonstrated the ability to make quantum chips. In simple terms, there are eight different ways, or so-called physical systems, to make these devices—photon; coherent state of light; electrons; nucleus; optical lattices; Josephson junction; singly-charge quantum dot pair; and quantum dots.
“Each has their own advantages and disadvantages,” Steffen said. “For example, your qubit could be an ion. Ions have incredibly long lifetimes, but they really don’t want to interact with each other. So it’s challenging to couple ions.”
IBM is pursuing the Josephson junction route using superconducting technology. “It’s straightforward,” he said. “We could probably manufacture a chip right now that has 100 or even 1,000 qubits coupled together. But the intrinsic performance is nowhere near what it is for ions.”
The technology is different than D-Wave’s qubit. IBM’s 3-qubit device consists of two components—a capacitor and a Josephson junction. The Josephson junction includes a thin insulating layer, which is sandwiched by two superconducting metals. The Josephson junction material is a combination of aluminum and aluminum oxide. “You can think of that as an inductor. The capacitors are used to form small quantum harmonic resonators. So basically, it’s a non-linear quantum resonator. In our case, that’s a quantum bit,” he said.
IBM’s devices can retain their quantum states up to 150 microseconds. The device itself does not require leading-edge lithography. The dimensions of the device are about 200 by 400 micrometers on a side. On the fabrication side, IBM’s devices are made using 300mm wafers. “We use standard tools,” he said. “The techniques are compatible, and the same, as the ones we use in silicon fabrication.”
Meanwhile, the QCCT in Australia is pursuing a different avenue based on two technologies—electron and nucleus. The group has devised a qubit-based device that consists of two gates—an A-gate and J-gate. The A-gate controls the interaction between a nuclear spin qubit and the electron spin. The J-gate controls the exchange interaction between the spins.
“These don’t work anything like devices that we are used to dealing with,” said NRI’s Theis, who is on assignment from IBM. “(The QCCT) has demonstrated that it is possible to make devices. I didn’t say it’s economically viable. The tool they are using to place dopants in those structures is much better than atomic bonding. They use phosphorous as dopants.”
The QCCT has developed two process flows. In one flow, the group can devise donor-based qubits using ion implantation. “This is taking conventional ion implantation and reducing it down to the single electron level,” said QCCT’s Simmons. “We take a resist structure and we put PMMA down. Then, we open these guide holes and do a blanket implant.”
With the process, the group has demonstrated MOS-based devices with a gate-controlled electron channel. The device also consists of phosphorous ions. “When we put a current through it, we create a magnetic field,” she said.
In the second flow, the group has devised an atomic fabrication process. “The other scheme we are using is putting the phosphorous atoms (in the device) with a scanning tunneling microscope (STM),” she said. “We first have to place registration markers into the silicon crystals. Then, we put down a layer of hydrogen on the surface, which acts as an atomic-scale mask.”
Then, the STM etches holes in the resist, exposing the silicon underneath the structure. Phosphine gas is applied to the structure. “We then go back with our STM tip and image the pattern we made on the surface. Using registration markers, we make contacts on the device,” she said.
Meanwhile, in the latest breakthrough, the University of Bristol has generated and manipulated single particles of photons on a quantum chip. Researchers have developed a silicon-on-insulator (SOI) device that combines two four-wave-mixing sources in an interferometer with a reconfigurable phase shifter.
Clearly, quantum computing is still in the early stages. When these systems finally appear in the market, they could solve several problems in the scientific field. But there are also some troubling privacy and data security issues. Still, the race is on in the arena, which, in turn, will fuel philosophical debates for years to come.