The world's largest maker of quantum computers, Canada's D-Wave Systems Inc., recently announced the Pegasus generation of its quantum computers, featuring 2.5 times the qubits (more than 5,000) than its predecessor, as well as the elimination of a major stumbling block to commercialization by directly connecting each of those qubits to three times as many nearby qubits as its previous generation, the Chimera.
Analysts are predicting that Pegasus will advance quantum applications down the technology lifetime exponential growth curve.
Exemplary applications include Denso Corp.'s automated guided vehicles without collision, T-QARD's factory optimization collaboration with Denso, Volkswagen's intelligent traffic management application, and Los Alamos National Laboratory's recent mathematical breakthrough showing how to use D-Wave's quantum annealing architecture to perform quadratic unconstrained binary optimization in place of supercomputer floating point operations.
"D-Wave's Pegasus opens up new application horizons," said Bob Sorensen, chief analyst for quantum computing at Hyperion Research. "By adding over twice as many qubits and three times as many interconnections, the quantum annealing operations performed by Pegasus can be adapted to many other applications besides optimization; from machine learning to financial portfolio risk-assessment."
"Quantum annealing" describes the hardware architecture used by all D-Wave processors. Quantum annealing sidesteps the need for extensive error correction, a problem that plagues the gate-level implementations of Google, IBM, Microsoft, Rigetti, and Xanadu, limiting them to fewer than 70 qubits. Only Fujitsu has also adopted quantum annealing, albeit in a 1024-bit digital chip that only emulates a quantum computer.
According to Sorensen, D-Wave's Pegasus topology takes it very close to the threshold of commercial growth by virtue of its vastly improved interconnection matrix, in which each qubit is connected to 15 nearby qubits. "Pegasus bridges the gap between quantum computing and real world of applications," said Sorensen.
The previous generation, Chimera quantum computers, only made direct interconnections between a qubit and five adjacent qubits, forcing programmers to "waste" some of its 2,000 qubits per chip as makeshift interconnections. The new arrangement will allow all 5,000 qubits to be used for equation variables, enabling complex real-world problems to be solved, according to Sorensen.
"For logistics applications, such as optimally routing taxis to riders and their destinations, Pegasus can now handle far more taxis than before," said Mark Johnson, vice president of processor design and development of quantum products at D-Wave. "Due to its higher connectivity architecture, equation variables can be represented with far fewer qubits. New applications can be taken closer to a positive return-on-investment [ROI], including route scheduling, financial portfolio optimization, and machine learning."
D-Wave claims that even though its quantum annealing needs no error correction, Pegasus does have improvements in precision that will make a big difference by shortening run times. Quantum annealing works in a manner similar to semiconductor or metallurgical annealing. It's an iterative process, and each iteration of the annealing process yields a more precise result, so greater precision at each step greatly decreases the overall run time for a given level of precision.
"The increased number of qubits and interconnections is interesting, but the big deal is Pegasus' higher precision," said Matthew Brisse, research vice president at Gartner Inc. "There are already over 100 [Chimera] applications on GitHub, but now programmers can accelerate those and their own algorithms by shifting to the new Pegasus topology."
According to Brisse, it will take five years or more before true quantum supremacy (the potential ability of quantum computing devices to solve problems that classical computers cannot) will be demonstrated on not just scientific, but also on everyday commercial applications.
The software development environment created by D-Wave, called Leap, is already educating programmers on how to best make the transition to quantum, without needing to learn the underlying physics. Programmer efforts are also being spurred by D-Wave's Ocean code library with standardized Python and C++ application programmer interfaces (APIs) plus Jupyter Notebooks, which allows interactive changes that instantly update results.
Brisse and Sorensen also praised D-Wave for directly responding to its user base when crafting the new Pegasus topology. Most D-Wave users run their quantum algorithms in the cloud, rather than purchasing the company's hardware. Luckily, D-Wave has unusually patient funding sources, since it is still getting their full support on its 20th birthday.
D-Wave says it was already leading the world in the sophistication of its superconducting process technology, but claims to have extended that lead with a newly enhanced process technology developed with SkyWater Technology, its microchip fabrication foundry. Not only does its current process offer niobium features below 240 nanometers, but also lowers its on-chip noise level, which for a quantum computer translates to longer coherence times—the length of time before data values collapse from a quantum superposition of states into a digital one or zero state. Thus besides enabling more quantum variables to tackle larger problems, according to D-Wave's Johnson, Pegasus' lower noise also enables the solution of more complex problems that require longer coherence times to solve.
Finally, Pegasus directly confronts how classical computers and quantum computers work together as companions, rather than as competitors. According to Johnson, hybrid configurations of classical and quantum computers will off-load to Pegasus only the parts of a problem that run better on quantum computers, while retaining a companion classical computer to run the rest of an application's algorithms. This hybrid architecture has been built into the latest revisions of Leap and its Ocean algorithm library to simplify quantum programming, as well as to demonstrate the best practices needed to perform hybrid classical-quantum computing.
"We really believe that successful customer solutions will be hybrids, so new operating software enhances hybrid configurations with lower latency and new options for scheduling problems," said Johnson. "In fact, D-Wave is the only company with real-time qubit scheduling, as well as block-of-time scheduling."
Real-time qubit scheduling allows algorithms to split up classical and quantum operations on a fine-scale, step by step, while block-of-time scheduling permits quantum and/or classical operations to be continuously executed without the overhead of switching modes.
Michael Cusumano at the Massachusetts Institute of Technology (MIT), who is familiar with programming the Chimera generation, says, "D-Wave's latest machine and programming environment sounds like real progress, and should be exciting for software developers working in the D-Wave ecosystem. Clearly, there are some significant technical advances, and the latest technology may well be closer to a general-purpose quantum computer."
D-Wave's Hybrid, for marrying an in-house classical computer to a cloud-based quantum computer, is available on GitHub, as are many of the Pegasus components. The rest of the Pegasus offerings will be rolled out by the company over the next 18 months, according to Johnson.
R. Colin Johnson is a Kyoto Prize Fellow who has worked as a technology journalist for two decades.