Using Cryogenic CMOS Control Electronics To Enable A Two-Qubit Cross-Resonance Gate

TL;DR

This paper demonstrates a CMOS-based control chip operating at 4K that successfully generates control signals for a two-qubit cross-resonance gate, achieving low error rates and providing insights into drive-induced errors in quantum computing.

Contribution

It introduces a 14nm FinFET CMOS ASIC for qubit control, integrating control electronics with quantum hardware at cryogenic temperatures, and characterizes its performance and error sources.

Findings

Achieved two-qubit gate with 1.4% error rate

Demonstrated control electronics operate at 4K with 23 mW power

Identified and modeled drive-induced Z-errors

Abstract

Qubit control electronics composed of CMOS circuits are of critical interest for next generation quantum computing systems. A CMOS-based application specific integrated circuit (ASIC) fabricated in 14nm FinFET technology was used to generate and sequence qubit control waveforms and demonstrate a two-qubit cross resonance gate between fixed frequency transmons. The controller was thermally anchored to the T = 4K stage of a dilution refrigerator and the measured power was 23 mW per qubit under active control. The chip generated single--side banded output frequencies between 4.5 and 5.5 GHz with a maximum power output of -18 dBm. Randomized benchmarking (RB) experiments revealed an average number of 1.71 instructions per Clifford (IPC) for single-qubit gates, and 17.51 IPC for two-qubit gates. A single-qubit error per gate of $\epsilon_{\text{1Q}}$=8e-4 and two-qubit error per gate of $\epsilon_\text{2Q}$=1.4e-2 is shown. A drive-induced Z-rotation is observed by way of a rotary echo experiment; this observation is consistent with expected qubit behavior given measured excess local oscillator (LO) leakage from the CMOS chip. The effect of spurious drive induced Z-errors is numerically evaluated with a two-qubit model Hamiltonian, and shown to be in good agreement with measured RB data. The modeling results suggest the Z-error varies linearly with pulse amplitude.