Reporting world-class qubit fidelity is one thing. Explaining precisely what was measured, how, and why it matters is another. In this post, I want to go beyond the headline number — 99%+ two-qubit gate fidelity on a 10-qubit germanium array — and explain the engineering and physics that got us there, and why achieving this on an array (rather than an isolated pair of qubits) is the result that matters for scalable quantum computing.
What Is Gate Fidelity?
Gate fidelity measures how close a real quantum gate operation is to the ideal unitary operation it's intended to implement. A fidelity of 99% means that, on average, 1% of the time the gate operation produces an output state that differs from the ideal output. The standard measurement technique is randomized benchmarking (RB), which applies long sequences of random gate operations and measures the decay of the survival probability with sequence length. The decay rate is directly related to the average gate error rate.
For two-qubit gates, the relevant fidelity is the average gate fidelity over all possible input states and all pairs of adjacent qubits in the array. This is a more stringent and more meaningful metric than the fidelity of a single, specially optimized gate on a single qubit pair. A processor where every gate pair in a 10-qubit array exceeds 99% fidelity is a qualitatively different beast from a processor where one pair of qubits achieves 99% while others are at 95%.
The Two-Qubit Gate in Germanium: Exchange Interaction
In germanium spin qubits, the two-qubit gate is implemented via the exchange interaction — the quantum mechanical coupling between the spins of holes in neighboring quantum dots. The exchange interaction arises from the overlap of the quantum dot wavefunctions when the barrier between them is lowered. By pulsing the barrier gate voltage, we can switch the exchange coupling on and off rapidly (in tens of nanoseconds), implementing controlled rotations of the two-qubit state.
The exchange interaction is a natural consequence of the quantum dot geometry and requires no additional microwave drives or resonators — just the standard DC gate electrodes that define the quantum dots in the first place. This simplicity is a major advantage of the spin qubit approach compared to superconducting qubits, which require carefully tuned microwave couplings between transmon elements.
Why 99%+ Across the Array Is Hard
Achieving 99% fidelity on a single qubit pair in isolation is primarily a calibration and pulse optimization problem. Given enough time to characterize the device and optimize the pulse shapes, most reasonable qubit platforms can demonstrate excellent two-qubit fidelity in ideal conditions. The hard problem is doing it simultaneously across all qubit pairs in a multi-qubit array, under the following constraints:
First, crosstalk: changing the exchange coupling between qubits 3 and 4 also shifts the electrostatic environment of qubits 2 and 5. Each gate operation perturbs the qubit frequencies of neighboring qubits, causing errors unless these perturbations are carefully compensated. Managing crosstalk in a 10-qubit array requires solving a coupled optimization problem with tens of parameters simultaneously.
Second, charge noise: the exchange coupling is exponentially sensitive to the gate voltages that control the inter-dot barrier. Low-frequency charge noise shifts these voltages randomly, causing random variations in the exchange coupling strength and therefore random errors in gate operations. Reducing the impact of charge noise requires a combination of improved material interfaces (less noise) and optimized pulse shapes (noise-robust gates).
Third, spatial variation: no two quantum dots in a fabricated array are exactly identical. Gate offsets, qubit frequencies, and coupling strengths vary across the array. High-fidelity operation requires individual calibration of every qubit and every qubit pair — a process that must be automated to be practical at scale.
How We Solved It
Our 99%+ result was achieved through three main advances. First, we improved the germanium quantum well quality and gate oxide interfaces, reducing charge noise at low frequencies by roughly a factor of three compared to our previous generation of devices. Second, we developed and implemented DRAG (Derivative Removal via Adiabatic Gate) pulse shaping for two-qubit gates, which suppresses errors from charge noise and unwanted higher-order transitions. Third, we built an automated calibration and benchmarking pipeline that can characterize all qubits and qubit pairs in the array systematically and efficiently.
The combination of better materials, smarter pulses, and systematic calibration is what moved us from competitive-but-not-leading fidelity to world-class fidelity across the full array. No single trick got us there — it was the disciplined application of best practices at every level simultaneously.
This approach — systematic, engineering-disciplined, across the full system rather than on cherry-picked subsystems — is how we intend to continue improving as we scale to larger arrays. The 99%+ result is not a ceiling; it is a foundation.