IBM Offers Quantum Error Suppression Out Of The Box

The error-prone nature of today’s quantum computers can make doing any useful computation on them a major headache. IBM has announced that, as of this past week, they have integrated error suppression technology from Q-CTRL into IBM cloud quantum services, letting users slash error rates by simply flicking a switch.

Computers that harness the unusual properties of quantum mechanics will ultimately be capable of computational feats beyond even the most powerful supercomputers. But the quantum states that make this possible are incredibly fragile and susceptible to noise, which means carrying out operations before they are overwhelmed by errors is a significant challenge.

“It’s a bit of a dirty secret in our sector that the typical user experience is very rarely at the limit of what the hardware could provide.”
—Michael Biercuk, CEO, Q-CTRL

It’s widely accepted that large-scale quantum computers will require some form of error correction. The leading schemes involve spreading information over a large number of physical qubits to create more robust “logical qubits”. But this can require as many as a thousand physical qubits for each logical one. Given that today’s largest processors feature just hundreds of qubits, error corrected quantum computing is still a distant prospect.

In the meantime, the start-up Q-CTRL—based in Sydney, Australia—says the best way to tame unruly, near-term quantum processors is “error suppression,” which involves altering how you operate the underlying hardware to reduce the likelihood of errors. Using a combination of techniques, the company says its software can boost the chances of an algorithm running successfully by several orders of magnitude. And now, IBM has integrated the technology into its quantum cloud offerings.

“It’s a bit of a dirty secret in our sector that the typical user experience is very rarely at the limit of what the hardware could provide,” says Q-CTRL CEO and founder Michael Biercuk. “That’s because the performance is effectively buried by all the sources of noise and interference and error. We, through our performance management solution, suppress that error in a way that allows a user to immediately access just about the best the hardware can theoretically deliver.”

IBM’s cloud quantum computing servers now use an error-correction protocol developed by the Sydney, Australia-based Q-CTRL, which they report provides many times more reliable results in quantum computations, compared to baseline systems without the protocol in place. Q-CTRL, adapted from Fig. 3, https://doi.org/10.1103/PhysRevApplied.20.024034

The company’s software requires no configuration by IBM’s end users. Customers accessing Big Blue’s quantum hardware over the cloud via its Pay-As-You-Go plan will simply see a “performance management” option that can be toggled on and off. Flicking the switch will engage an automated set of several different software modules that run in the background to optimize the way the user’s algorithm runs on the hardware.

According to Biercuk, Q-CTRL’s quantum compiler mathematically optimizes the number of logic gates required to run an algorithm before subjecting this minimal circuit to several further error suppressing steps. For a start, the gates are mapped onto the hardware’s qubits in such a way as to avoid the most error-prone layouts, based on a catalog of pre-performed test measurements. The circuit is also interleaved with operations that use a technique known as “dynamical decoupling,” in which qubits are subjected to control pulses designed to cancel out the effects of cross-talk from other nearby qubits.

Separately, Biercuk says, AI designed by Q-CTRL regularly redefines the machine language used to implement circuits on the hardware. Every six to twelve hours, he says, the AI runs through multiple test circuits to check how various potential gate implementations contribute to errors. The results are then compiled in a lookup table that the compiler uses to build the circuit.

“We break the link between the underlying quantum processor ... and what the application-focused end-user programs. That’s a huge change.”
—Michael Biercuk, CEO, Q-CTRL

Finally, once the circuit has run, the software then carries out a final post-processing step on the results designed to catch measurement errors. “One of the ways that quantum computers fail is that the algorithm actually runs correctly, but when you look at the answer to read it out, that process is faulty,” says Biercuk, who is also a professor of quantum physics at The University of Sydney. Q-CTRL’s software uses a neural network to learn patterns in the way measurement errors occur in the hardware. The system then uses these patterns to offset any errors in the readout.

With all these error-mitigation efforts working in concert, these different modules boost the chances that an algorithm will run successfully—which is not a sure fire thing when it comes to quantum computers. In peer-reviewed research published in Physical Review Applied in August, the company tested their error suppression technology out on several popular quantum algorithms and showed that they could boost performance by as much as 1000 times. (The quantum circuit Q-CTRL tested their error-correction protocols on in the August paper was more limited in size. However, contacted by Spectrum about the discrepancy between the size of the quantum circuits described in the graph above and those depicted in the August paper, Biercuk emailed back, “It wasn’t published in the previous paper, because at the time those larger devices didn’t exist yet.”)

The company’s software works with any kind of quantum computing hardware, says Biercuk, be that trapped-ions, superconducting qubits or cold atoms. And while configuring the software for a particular processor takes some time and effort, there is no extra computational cost for the user at runtime.

Biercuk thinks an out-of-the-box solution for error suppression could be a boon for many of the companies interested in quantum computing. Typically, they have focused on algorithm development, but error-suppression requires expertise in low-level hardware manipulation.

“It’s a little bit like asking a web developer who’s building Salesforce or Facebook, can you please start programming at the level of the voltages on the transistors,” says Biercuk. “We break the link between the underlying quantum processor and the way it has to be manipulated, and what the application-focused end-user programs. That’s a huge change.”

Quantum hardware is still some way from being able to compete with classical computers on practical problems, admits Biercuk, but a number of companies have been testing out the technology ahead of the integration with IBM.

“We have previously explored Q-CTRL’s performance management capabilities and were impressed by the order of magnitude improvement seen across both the inverse quantum Fourier transform and quantum phase estimation,” Julian van Velzen, CTIO and Head of Quantum Lab at Capgemini in Utrecht, the Netherlands, said in a statement. “With this technology natively embedded within IBM Quantum services, we can get more value from current hardware and push our applications further.”