If gate fidelity is the metric that gets the most attention in quantum computing benchmarks, coherence time is the one that tells you the most about the underlying quality of the hardware. Coherence time is, in essence, how long a qubit can hold a quantum state before it deteriorates into random noise. Everything else — gate speed, error correction performance, algorithm depth — ultimately depends on it.
What Coherence Time Measures
There are actually several distinct coherence times that characterize a qubit, and they measure different aspects of the same underlying phenomenon — the interaction between a qubit and its environment.
T1 (energy relaxation time) measures how long it takes for a qubit in its excited state to spontaneously decay to its ground state. This is analogous to radioactive decay: the qubit has a certain probability per unit time of losing its excitation energy to the environment. T1 sets a hard upper limit on how long a qubit can maintain an arbitrary quantum state.
T2 (dephasing or decoherence time) measures how long a qubit can maintain a coherent superposition of its 0 and 1 states. T2 captures an additional source of error beyond energy relaxation: phase noise. Phase noise can arise from fluctuating magnetic fields, charge noise in the gate electrodes, nuclear spin fluctuations, or temperature variations. T2 is always less than or equal to 2*T1, but in practice it is often much shorter than the T1 limit, dominated by phase noise sources.
For gate operations to be reliable, the gate time must be much shorter than the relevant coherence time. If your two-qubit gate takes 100 nanoseconds and your T2 is 1 microsecond, you can execute about 10 gates before the qubit loses coherence — not enough for any useful computation. Modern quantum processors aim for T2/gate_time ratios of at least 1,000.
What Limits Coherence in Germanium Devices
In germanium hole-spin qubits, the dominant noise sources at cryogenic temperatures are charge noise and (at higher temperatures) phonon noise. Charge noise arises from fluctuating charges trapped in defects at material interfaces — particularly at the germanium/oxide interface beneath the gate electrodes. These fluctuating charges create small, random electric fields that shift the energy levels of the quantum dot and cause dephasing.
This is one of the primary reasons why germanium quantum well quality matters so much. Defects at the Ge/SiGe interface or in the gate dielectric act as noise sources that directly limit T2. A significant fraction of our device engineering effort goes into minimizing interface trap densities through careful heterostructure design, optimized oxidation conditions, and surface passivation.
A second noise source in germanium, as in all spin qubit platforms, is hyperfine coupling to nuclear spins. Germanium is fortunate here: naturally occurring germanium is about 92% spinless isotopes (Ge-70, Ge-72, Ge-74, Ge-76 have zero nuclear spin), with only about 7.8% Ge-73 (nuclear spin 9/2). This is in contrast to silicon, where natural silicon is also predominantly spinless, and to gallium arsenide, which has 100% nuclear spin and therefore terrible spin qubit coherence without isotopic purification.
Further isotopic purification of germanium — enriching to greater than 99% spinless isotopes — is expected to improve T2 significantly and is a near-term research direction for several groups including ours.
Engineering Longer Coherence: Our Approach
Over the past three years, we have systematically improved T2 in our germanium devices through a combination of heterostructure optimization, gate stack engineering, and dynamical decoupling pulse sequences.
Dynamical decoupling deserves explanation: even if you can't eliminate noise sources, you can design pulse sequences that make the qubit insensitive to low-frequency noise components. By applying a carefully timed series of pi pulses that periodically flip the qubit state, low-frequency noise contributions (which tend to be the largest) average out to zero over the sequence duration. This can extend effective T2 by factors of 10-100 compared to the bare Hahn echo T2.
The combination of improved material quality and optimized pulse engineering has enabled us to achieve coherence times that, combined with our fast gate operations (two-qubit gates below 100 ns), give T2/gate_time ratios well above 1,000 — firmly in the regime where high-fidelity gate operations are achievable and quantum error correction becomes tractable.
Why Coherence Time Matters for the Future
As quantum processors scale to larger qubit counts, maintaining coherence across all qubits simultaneously becomes increasingly challenging. Each additional qubit adds potential noise sources, and controlling a large array requires more complex electronics that can themselves introduce noise. This makes it imperative that the individual qubit quality — including coherence time — be as high as possible before attempting to scale.
We view long coherence time not as a final goal but as the foundation on which everything else is built. You cannot engineer good gate fidelity without good coherence. You cannot implement effective quantum error correction without gates that are fast relative to coherence time. And you cannot scale to thousands of qubits without the engineering discipline that achieving long coherence time demands. It is a proxy for the overall quality of the hardware platform — and it is a number we intend to keep improving.