By Blake Johnson, VP Quantum Engineering
Scaling near-term quantum computers is about more than adding qubits. Qubit performance must also improve proportionally, especially for noisy, intermediate-scale quantum (NISQ) devices that operate without the benefit of quantum error correction. Specifically, today’s quantum processors are limited by the error rates of 2-qubit gates. Tackling this challenge is a primary focus of our work at Rigetti as we build toward our 128-qubit processor.
Qubits are sensitive to interactions with their environment, with any noise affecting the quality of a computation. Building high-quality logic gates is difficult given the dual goals of having to both isolate and control qubits. We isolate qubits to protect them from noise, while simultaneously coupling those qubits to control electronics in order to enact gates. Recently, we developed a better way to isolate and control the qubits at the same time, resulting in a two-qubit gate with over 99% fidelity.
The gate we used works with qubits connected via passive circuit elements, such as capacitors. The qubits are normally operated at different frequencies, where the interaction from the capacitor is effectively ‘off’. To turn the gate on, we send a control signal to one of the qubits to modulate its frequency. Like in an FM radio, when we shake the qubit frequency back and forth sidebands, or peaks, appear in the qubit spectrum at multiples of the modulation frequency. By aligning one of the sidebands of the tunable qubit with a neighboring qubit, we allow energy to swap between the qubits. Through careful calibration of the control signal, this interaction becomes a controlled-Z (CZ) gate.
We previously demonstrated this method of enacting a CZ gate last year. However, our prior processors were limited to at most 95% gate fidelity, with a wider range of performance on the largest devices. The problem was that modulation of the qubit frequency to enable the gate introduced noise that caused additional decoherence in the qubits. Last year, our theory team discovered a special operating condition called an AC sweet spot where the qubits would be less sensitive to noise from the modulating control signal. The theorists warned us, though, that we would need to reduce the noise of our control electronics to observe the full benefit.
After many months of effort, we were able to build better electronics and redesign our chips to operate gates at the AC sweet spot. Our previous control system was based upon an off-the-shelf component that was designed as a software-defined radio. It was sufficiently flexible to adapt it to quantum computing, but it was also quite noisy, since state-of-the-art noise performance is not necessary for a typical radio. Our in-house control system not only produces cleaner signals, but also makes the system programmable in the true sense of the word, as it allows for branching behavior (i.e. “if-then-else”) in programs from qubit measurements. This programmability is at the heart of the qubit state initialization (reset) protocol used in the Quantum Cloud Services™ (QCS) platform.
In a first proof of concept, we studied a pair of qubits on an 8-qubit chip modeled after a section of our Aspen lattice. Our work clearly shows the emergence of the predicted sweet spot. When operating a CZ gate at that special point, we see gate fidelities as high as 99.2%. This is comparable to the best 2-qubit gate fidelities reported for other superconducting processors.
We deliberately chose to mimic the essential features of our larger Aspen lattices for this work, so that improvements made on smaller devices could be quickly ported into larger ones. The devices deployed to QCS this year will feature steady increases in performance as we dial-in device properties to consistently produce good performance. The device used in this demo was produced with a legacy fabrication process that did not incorporate lessons learned from our coherence study. Combining these two advances should yield even greater performance gains in the future.
The better our qubits interact with each other, the more complex computations we can do on near-term devices and the more confident we can be in the solutions we get. The approach described here is one of several we are taking at Rigetti to consistently and iteratively improve the performance of our chips.
S. Hong, A. Papageorge, and P. Sivarajah, et al (2019). “Demonstration of a Parametrically-Activated Entangling Gate Protected from Flux Noise”. https://arxiv.org/abs/1901.08035