Qubit Transistors Achieve Error Correction Benchmark

A team in Australia recently made a major advance in their metal-oxide-semiconductor (or MOS-based) quantum computer. They showed that two-qubit gates (logical operations involving two or more quantum bits, or qubits) were performed without error 99% of the time. This number is significant because it is the benchmark needed to perform error correction, which is believed to be necessary for building large-scale quantum computers. Moreover, these MOS-based quantum computers are compatible with existing CMOS technology, which should make it easier to fabricate large numbers of qubits on a single chip than with other technologies.

“Getting above 99 percent is important because that’s what many consider the threshold for error correction. If you’re below 99 percent fidelity, it doesn’t matter what you do in terms of error correction,” says Yuval Boger, chief commercial officer at quantum computing company QuEra, who was not involved in the work. “You can’t correct errors faster than they accumulate.”

There are many competing platforms in the race to build a useful quantum computer. IBM, Google, and others are building machines with superconducting qubits. Quantinuum and IonQ use individually trapped ions. QuEra and Atom Computing use neutrally charged atoms. Xanadu and PsiQuantum are betting on photons. The list goes on.

In the new results, a collaboration between the University of New South Wales (UNSW) and Sydney-based startup Diraq, along with contributors from Japan, Germany, Canada and the US, has taken a different approach, trapping single electrons in MOS devices. “What we’re trying to do is to create qubits that are as close as possible to a traditional transistor,” says Tuomo Tantú, a UNSW researcher who led the effort.

Qubits that work like transistors

These qubits are actually very similar to regular transistors, gated so that there is only one electron in the channel. The biggest advantage of this approach is that it can be manufactured using existing CMOS technology, theoretically scalable to millions of qubits on a single chip. Another advantage is that MOS qubits can be integrated into standard transistors and chips, simplifying inputs, outputs and control. Diraq CEO Andrew Dzurak said.

However, a downside to this approach is that MOS qubits have historically suffered from significant device-to-device variability, resulting in significant noise in the qubits.

“The sensitivity of a (MOS) qubit will be higher than that of a transistor, because in a transistor you still have 20, 30, 40 electrons carrying current. In a qubit device, you actually have just one electron,” says Ravi Pilarisetti, a senior device engineer for quantum hardware at Intel, who was not involved in the work.

The team’s results not only demonstrated 99 percent accuracy in the two-qubit gates of their test device, but also helped them better understand the source of device-to-device variability. The team tested three devices, each with three qubits. In addition to measuring error rates, they also conducted extensive studies to determine the underlying physical mechanisms contributing to the noise.

Researchers found that one source of noise was isotope impurities in the silicon layer, and controlling these greatly reduced the circuit complexity required to run the device. The next major source of noise was small variations in the electric field, which were probably due to imperfections in the oxide layer of the device. Tanttu says that moving from a lab cleanroom to a casting environment could potentially improve this.

“It’s a great result and a great step forward, and I think it’s setting the community in the right direction in terms of thinking more long-term about the scalability path as opposed to thinking less about individual devices or demonstrating something on an individual device,” says Filarisetti.

The challenge now is to scale these devices to more qubits. One challenge in scaling is the number of input/output channels required. Intel’s quantum team, which is pursuing a similar technology, recently pioneered a chip called the Pando Tree to address this problem. The Pando Tree sits on the same substrate as the quantum processor, allowing faster input/output to the qubits. The Intel team hopes to use this to scale to thousands of qubits. “A big part of our approach is thinking about how to make the qubit processor look more like a modern CPU,” Pillarisetty says.

Likewise, Diraq CEO Dzurak says his team plans to scale the technology to thousands of qubits in the near future, thanks to a recently announced partnership with Global Foundries. “Together with Global Foundries, we have designed a chip that will have thousands of these[MOS qubits]and they will be interconnected using classical transistor circuits that we have designed. This is unprecedented in the world of quantum computing,” says Dzurak.