Choosing a qubit species for a neutral-atom platform is one of the earliest and most consequential decisions in designing the system. The choice determines which laser wavelengths you need, what transitions you use for cooling and detection, how you encode the qubit, and what the dominant decoherence mechanisms will be. Different groups have made different choices: rubidium-87 and cesium-133 have large hyperfine splittings that make qubit encoding straightforward; ytterbium-171 and ytterbium-174 have optical clock transitions in fermionic and bosonic isotopes respectively; strontium-87 and strontium-88 represent the two ends of the same element.
We chose strontium-88. This post explains the physics reasoning behind that choice, which also illuminates the tradeoffs different neutral-atom platforms are navigating.
The clock transition: why it matters for a qubit
The qubit in our system is encoded in the ground state and the metastable excited state of the 1S0 to 3P0 clock transition at 698 nm. This transition is called a "clock transition" because it is the basis for strontium optical lattice clocks, which currently represent the most accurate frequency standards ever built (fractional uncertainties below 10^-18). The properties that make it useful for clocks also make it useful for qubits.
The 3P0 state is metastable: the transition to the ground state is forbidden by angular momentum selection rules to first order. For strontium-88, which has zero nuclear spin (a bosonic isotope), the 3P0 state has J=0 and the 1S0 ground state has J=0, making the transition a J=0 to J=0 transition that is strictly forbidden for electric dipole radiation and only weakly allowed through a higher-order mixing mechanism. The lifetime of the 3P0 state is approximately 160 seconds. This extremely long lifetime is the qubit's coherence time in the absence of other decoherence sources. In practice, coherence is limited by other factors (laser frequency noise, trap light shifts, stray magnetic fields) long before the natural lifetime matters, but the long natural lifetime provides enormous headroom that alkali atoms with microsecond excited-state lifetimes do not have.
Zero nuclear spin and why it simplifies things
Strontium-88 has zero nuclear spin. This means the ground state has no hyperfine structure: there is only one ground state sublevel. For comparison, rubidium-87 has a nuclear spin of 3/2, giving the ground state a hyperfine structure with F=1 and F=2 manifolds separated by about 6.8 GHz. This separation is useful for qubit encoding (the two hyperfine levels are the qubit states) and for microwave addressing, but it also means the atom responds to magnetic field fluctuations differently in the two hyperfine states, and magnetic field noise couples to the qubit coherence.
With strontium-88's zero nuclear spin, there is no hyperfine structure to worry about. The 1S0 to 3P0 qubit transition is purely an optical transition between two J=0 states. J=0 states have zero electronic angular momentum, which means to first order neither state has a magnetic moment. The qubit frequency is therefore first-order insensitive to magnetic field fluctuations. This is a significant coherence advantage in lab environments where magnetic field fluctuations at power-line harmonics (50 Hz or 60 Hz) are difficult to eliminate entirely.
The tradeoff is that without hyperfine structure, you cannot use microwave addressing for single-qubit gates. All gate operations on strontium-88 qubits use optical pulses at 698 nm. This requires a laser with very narrow linewidth (below 1 Hz is needed to drive coherent rotations with high fidelity), which is technically demanding but achievable with a cavity-stabilized laser locked to an ultralow expansion (ULE) glass reference cavity. We use a 698 nm laser with a cavity linewidth of approximately 0.5 Hz and short-term stability better than 1 Hz over 1-second intervals, which is sufficient for high-fidelity qubit rotations.
Magic-wavelength trapping
When an atom is held in an optical tweezer trap, the trap light creates an AC Stark shift on each atomic energy level. If the two qubit states (1S0 and 3P0) have different polarizabilities at the trap wavelength, the trap light shifts the qubit transition frequency in proportion to the local trap intensity. This is problematic because intensity variations across the trap (due to the non-uniform intensity profile of a Gaussian beam) create a distribution of qubit frequencies across the array. Atoms in different parts of the trap oscillate at different Larmor frequencies, and after some time the qubit coherence dephases due to this inhomogeneous broadening.
A magic wavelength is a specific trap wavelength at which the 1S0 and 3P0 states have equal polarizabilities, so the AC Stark shift is the same for both states, and the net shift to the qubit transition frequency is zero to first order. For strontium-88, a magic wavelength for the 1S0 to 3P0 transition exists at approximately 813 nm. Trapping at 813 nm eliminates the differential light shift to first order, removing a dominant source of inhomogeneous dephasing from the qubit frequency distribution. This is one of the key reasons strontium-88 is favored for optical clock experiments and also why it is advantageous for qubit applications.
In practice, the magic wavelength is not exactly magic for all spatial positions in the trap, because higher-order polarizability terms and the intensity gradient of the trap beam create residual light shifts. But the suppression of the first-order differential light shift is substantial. The residual inhomogeneous broadening from higher-order effects is on the order of a few hertz across a 100-site array, compared to kilohertz-scale inhomogeneous broadening that would be present without magic-wavelength trapping. This makes T2* of the array much better than would otherwise be achievable.
The Rydberg excitation pathway
Two-qubit gates in our system are implemented via the Rydberg blockade mechanism: two atoms are simultaneously excited to high-lying Rydberg states with principal quantum number n in the range 50-80, where they interact through the dipolar van der Waals interaction. The interaction energy for two atoms at separation R scales as C6/R^6, where C6 is a coefficient that grows as n^11. At our typical site separations (5-8 micrometers), and for Rydberg states in this range, the interaction energy is much larger than the Rydberg excitation linewidth, creating the blockade condition: only one atom in the pair can be excited to the Rydberg state at a time.
Exciting strontium-88 to a Rydberg state requires a two-photon process because the ionization potential is too high to reach efficiently with a single UV photon at practical power. The standard pathway in our system uses a first photon at 461 nm (the 1S0 to 1P1 transition) followed by a second photon at approximately 319 nm (from 1P1 to a high-n Rydberg state). An alternative pathway through the 3P1 intercombination line at 689 nm is also used in some groups; we use the 461 nm path because the larger linewidth of 1P1 (32 MHz) makes the intermediate state more accessible with less stringent laser frequency requirements, at the cost of higher spontaneous emission rate from the intermediate level during the excitation pulse.
A feature worth noting: because the qubit ground state is 1S0 and the qubit excited state is 3P0, the Rydberg excitation pathway from 1S0 does not address the 3P0 qubit excited state directly. This means Rydberg excitation drives transitions from the 1S0 qubit ground state, and the gate protocol uses the presence or absence of Rydberg excitation (conditional on the 1S0 occupation) to implement the controlled phase gate. The 3P0 qubit excited state is "dark" to the Rydberg excitation beam, which is operationally convenient for some gate protocols. The specific gate circuit that exploits this geometry is described in detail in our post on Rydberg blockade fidelity improvements.
What strontium-88 does not give you
Strontium-88 is not the right choice for every neutral-atom application. The bosonic isotope lacks nuclear spin, which means it cannot encode a hyperfine qubit. If you want a microwave-addressable qubit with a large frequency difference between states (important for certain qubit control architectures), strontium-87 (the fermionic isotope with nuclear spin 9/2) or rubidium-87 would be more appropriate. The all-optical qubit control required by strontium-88 demands narrow-linewidth lasers that are expensive and require careful maintenance.
The 461 nm laser for blue MOT operation and Rydberg excitation path is technically more demanding than equivalent infrared systems for alkali atoms. Getting a high-flux source of strontium atoms for the MOT is somewhat more work than for alkali atoms, because strontium requires oven temperatures above 400 C for adequate vapor pressure and the atom must be slowed from a high initial velocity. These are engineering challenges, not fundamental physics barriers, but they do represent upfront system complexity that some groups find preferable to avoid by working with alkali atoms.
The choice of strontium-88 reflects a bet that the coherence properties and magic-wavelength trapping advantages outweigh the added optical complexity. From where we stand having built and operated the system, that bet looks correct for the high-fidelity gate applications we target. We would not expect the same conclusion to hold universally across all neutral-atom applications.