Google over the last few years has thrown shade at today’s fastest supercomputers with dubious claims of achieving “quantum supremacy.” The tech giant may be backing away from that approach after its validity was questioned by researchers.
Last week, the company said it had created a stable quantum computer with mechanisms to make sure the system is robust. The company said its newest quantum system had error-correction techniques that can generate reliable output with fewer errors.
Google said the quantum system offered a combination of fewer errors and better performance than its previous systems. But the company also found out that it had to sacrifice a lot of quantum performance in order to bring stability to the system.
Unlike its past papers on quantum breakthroughs, Google did not claim quantum supremacy (or the more restrained “quantum advantage”), a concept in which quantum computers can significantly outperform classical systems, which are unable to solve similar problems in reasonable time.
Instead, Google’s latest research is along the lines of creating fault-tolerant systems that can withstand the worst of bumps thrown at it.
“For the first time ever, our quantum AI researchers have experimentally demonstrated that it’s possible to reduce errors by increasing the number of qubits,” said CEO Sundar Pichai, in a blog entry last week.
Quantum computers are notoriously unreliable as qubits can easily spin out of control, which can wreck calculations. Google is claiming the error-free system opens the door to build larger quantum systems with stable qubits that interact in a coherent fashion.
The reliability on Google’s quantum computer hangs on better error correction features. The 49-qubit system suppresses more quantum errors by scaling up the size of codes used to maintain predictable behavior. The hardware also has stabilizers to maintain system coherence and qubit stability.
Google’s breakthrough is analogous to high-reliability features like RAS – or reliability, availability, and serviceability – which made mainframe computers fault tolerant. Those advanced error-correction features, which were in chips like Intel’s Itanium, are now standard in x86 server chips.
Quantum companies so far have measured quantum progress by the number of qubits. But Google’s 49-qubit system outlines a shift in how the company is approaching its quantum research, with a focus on system stability.
Current quantum computers are seen as being functional but not as reliable as classical supercomputers to carry out massive computations. Major cloud providers and supercomputing hosts like Jülich Supercomputing Centre, are still exploring applications that are best suited for quantum computers.
Google in 2019 claimed quantum supremacy, which is a concept based on a 2012 paper by John Preskill of quantum computers outperforming classical computers, which are unable to do similar calculations in a reasonable amount of time.
Google’s controversial quantum supremacy paper in 2019 was based on a 54-qubit system called Sycamore, in which the qubits are arranged in a 2D array. Google said that its 54-qubit was able to solve specific problems in 200 seconds, which was significantly faster than the 10,000 years it would take classical supercomputers on the same tasks.
That claim was immediately challenged by IBM, which said that while Google’s achievement was impressive, it was “being broadly misinterpreted and causing ever growing amounts of confusion.” IBM’s Summit was considered the world’s fastest supercomputer at the time.
IBM later in 2019 went on to publish a paper that was aimed at disproving Google’s theory. The IBM paper pointed out that its Summit computer – with additional secondary storage – was able to achieve six times better performance than the one claimed in Google’s quantum supremacy paper, and solve problems in a reasonable amount of time.
Google’s claim of quantum supremacy was again challenged in 2022 by various researchers, including a group from Hebrew University of Jerusalem, which tried to investigate the tech giant’s quantum supremacy claims.
The researchers noted that Google’s closed-door experiments were flawed for reasons that included comparing its optimized quantum algorithms against older classical algorithms, which were slower. The researchers called for a new Google quantum-to-classical computing comparison to newer classical algorithms that could solve in a matter of seconds.
In 2018, Google also said that “the experimental demonstration of a quantum processor outperforming a supercomputer would be a watershed moment for our field, and remains one of our key objectives.” Google laid out in 2018 a plan to demonstrate a system “with 49 qubits, a circuit depth exceeding 40, and a two-qubit error below 0.5%.”
Google has realized that plan with its latest research breakthrough, but the company is staying far away from boasting about quantum supremacy. A paper on the research, which was published in Nature last week, makes no claims of the quantum computer outperforming the fastest supercomputers.
Instead, Google’s latest quantum breakthrough involves scaling up its system to 49-qubits while correcting more errors. The system has more surface code — which is code to detect and correct quantum errors – attached to the quantum bits.
The technique brings reliability and prevents qubits from spinning out of control as it goes into superposition. It also prevents the creation of an unstable environment in which calculations break down as system performance scales up.
“The latest quantum chip is the most powerful we have ever made. But it still needs to be thousands of times larger in order to solve all the practical applications that people are excited about. The problem is, as you make it bigger, you get more errors,” said Julian Kelly, director of quantum hardware at Google, in a video published on Youtube.
Early attempts to run the algorithm on the quantum error-corrected system yielded terrible performance, but modifications over months yielded better performance and reduced error rates.
But there was not much in terms of performance gains; the 49-qubit system setup only “narrowly” outperformed its average 17-qubit setup, the research paper said.
But performance aside, even the slightest reduction of errors in a larger qubit count opens the door to create larger quantum systems, the Google researchers said.
Google’s goal is to put out an “error free” quantum computer by 2029. The latest breakthrough represents a shift of Google’s quantum goals to focus on system stability, and away from claims of quantum supremacy.
Academic researchers have said there are many approaches to developing quantum computers, but the number of qubits will not matter if the system isn’t reliable.
“The question of the feasibility of powerful quantum computers beating classical super-HPC hinges on that it will be ultimately possible to perform quantum error correction,” said Göran Wendin, professor of quantum technology at the Chalmers University of Technology, in a research paper published on February 9.
IBM says the quantum advantage inflection point is nigh. The company’s roadmap includes a system called Flamingo in 2024 with more than 1,386 qubits, and a system called Kookaburra with more than 4,158 qubits.
Meanwhile, a team from China said it has already grabbed that ring, last year claiming quantum advantage for a 60-qubit system, which took 4.2 hours to solve a problem that a supercomputer would not be able to do in a reasonable amount of time. The paper noted the measurements included “continuously improved classical algorithms and hardwares,” and the “sampling task is about 6 orders of magnitude more difficult than that” of the 54-qubit system that Google used in its controversial 2019 paper.
Wendin said that “to create useful applications showing quantum advantage, it is necessary to scale up QPUs and related classical-quantum hybrid (HPC+QC) infrastructure.”
Public and hybrid cloud providers see an infrastructure emerging with quantum computers available as accelerators in conventional data-center environments. Dell offers access to an IonQ quantum computer in its computing infrastructure. All major cloud providers also offer access to experimental quantum systems.
Quantum computing researchers are also working on the concept of quantum-circuit cutting, in which the most demanding computing work goes to a quantum computer, and lower-priority work being offloaded to GPUs. Nvidia has a toolkit called QODA that can simulate quantum workloads on GPUs.
While cash-rich Google and IBM can afford to spend more time on research, the story is different for quantum hardware startups that are struggling financially. Startup Rigetti recently reset its quantum hardware roadmap to prioritize error-mitigating quantum systems.