PaQit: Energy-Runtime-Fidelity Co-Optimization for Neutral Atom Quantum Computers

AI summary

Overview

Research area: Systems-level modeling and scheduling for neutral-atom (Rydberg-atom) quantum computers, sitting at the intersection of quantum device physics, resource scheduling, and energy-aware computing.

Technical level: Intermediate to Advanced. The paper assumes familiarity with the Rydberg blockade, the van der Waals interaction, and basic scheduling/energy models, but the core argument is readable with an undergraduate physics or systems background.

Scope: The paper builds an analytical energy-and-runtime model for neutral-atom quantum computers, derives a fidelity-versus-qubit-packing relationship, and validates the resulting scheduler (PaQit) on simulations and real QuEra Aquila hardware.

What This Paper Is About

Neutral-atom quantum computers are known to be relatively low-power because they need no cryogenics, but nobody had formally connected system-level energy use, execution time, and computational fidelity into one model that a scheduler could act on. The authors show that because these machines are dominated by a large static power baseline, packing many circuit "shots" onto the array in parallel can cut both runtime and energy — except that packing atoms closer together creates unwanted Rydberg crosstalk that damages fidelity. PaQit turns that trade-off into a tunable policy: given a target fidelity, it computes how densely qubits can be packed and therefore how much time and energy the workload will cost.

Key Contributions

1. First-principles energy and runtime model. The authors derive a power law of the form P(n) = P₀ + αn^β and use it to derive analytical expressions for total runtime and energy under serial execution (Eqs. 4–5) and maximally parallel execution (Eqs. 6–8), establishing upper and lower bounds on both metrics.

2. Characterization of packing-induced fidelity degradation. They formulate fidelity as f(η) = f_max · exp[−κ(r_b/d₀)⁶η³], where η is packing efficiency, explicitly linking spatial density to interaction-driven error via the 1/r⁶ interaction.

3. PaQit, a fidelity-aware qubit packing framework. Inverting the fidelity model gives the maximum allowable packing efficiency η(f*) for a target fidelity (Eq. 13), which feeds into effective runtime (Eq. 14) and effective energy (Eq. 15) expressions, plus a formal optimization objective (Eq. 16) that minimizes energy subject to a fidelity constraint. The framework requires no hardware modification.

4. Validation on QuEra systems and real hardware. The model is checked against simulations of QuEra's analog Aquila system and digital Gemini system, and against real executions on Aquila via AWS Braket. The paper also states that spacing beyond 10 μm incurs negligible fidelity loss on Aquila, enabling efficient parallel execution.

Main Findings

• Static power dominates. QuEra Aquila consumes approximately 7 kW at 256 qubits and is projected to require only around 10 kW at 10,000 qubits; the calibration uses P₀ = 6.92 kW with α = 0.000308 kW/qubit and β = 1. Because P₀ ≫ P_dyn(n), energy becomes primarily a function of execution time rather than instantaneous qubit utilization.

• Parallelism is effectively "free" in energy but not in fidelity. Additional qubits cost little power, so amortizing the static baseline across concurrent shots lowers both runtime and total energy — until interaction-induced crosstalk becomes dominant.

• Packing efficiency falls steeply with fidelity target. Figure 3 shows η(f*) dropping sharply as target fidelity rises, with the decline highly sensitive to the normalized blockade ratio r_b/d₀.

• Runtime and energy rise with fidelity. Figures 4 and 5 show total runtime and energy increasing as fidelity targets tighten, because fewer circuit instances can run concurrently and shots must be partially or fully serialized. The paper describes the energy profile as convex, revealing an optimal operating point where marginal fidelity gains begin to incur disproportionate runtime and energy costs.

• Interaction physics, not qubit count, limits throughput. For larger r_b/d₀ ratios, even modest increases in fidelity requirements cause sharp reductions in parallelism; the authors conclude that increasing hardware scale alone is insufficient without sufficient spacing to control interaction errors.

• Dense packing degrades state-preparation fidelity in simulation. On a simulated Aquila array, fidelity stays near unity when atoms are well separated or only one atom is loaded, and degrades as lattice spacing shrinks and more atoms are packed in. The dense-packing regime probed corresponds to d₀ ≈ 4 μm, blockade distance r_b ~ 6–8 μm, and r_b/d₀ ~ 1.5–2, with Ω_max = 1.58×10⁷ rad/s and |Δ_max| = 1.25×10⁸ rad/s.

• The fidelity-aware scheduler reduces total energy while maintaining stable performance. The paper states that fidelity-aware packing can significantly reduce total energy, with predicted trends in close agreement with observed behavior on Aquila and Gemini.

• Circuit structure matters separately from packing. On simulated Gemini line qubits (N ∈ {2,…,9}), four circuit families — GHZ chain, disjoint Bell pairs, brickwork CZ, and fully connected CZ — are compared under one noise model, isolating how entangling density alone affects fidelity. Reported native Gemini fidelities exceed 99.7% for single-qubit gates and 99.2% for parallel two-qubit gates.

Methodology in Plain English

The authors start from the physics: atoms in a tweezer array are excited to Rydberg states, and any two excited atoms interact with strength C₆/r⁶. That interaction is what makes entangling gates work, but it also means two independent copies of a circuit placed too close together will interfere.

They write down a simple power model for the machine — a large constant baseline plus a small per-qubit term — and then compute total energy and time under two extremes: running one shot at a time, and packing the array as full as possible. Those give the worst-case and best-case bounds.

Next they add physics to the best case. Spacing is assumed to scale as d = d₀η^(−1/2) on a 2-D square lattice, so denser packing shrinks spacing, and fidelity is modeled as an exponential penalty in (r_b/d)⁶. Inverting that gives the densest packing allowed for any target fidelity, which plugs straight back into the runtime and energy formulas. The scheduler then just picks the density that minimizes energy subject to the fidelity requirement.

For validation they build a Python 3.11 discrete-event simulator using s = 1000 shots, n = 16 qubits, N = 256 total qubits, t = 16 μs, κ = 1.0, and f_max = 1.0, sweeping target fidelity from 0.8 to 1.0 and r_b/d₀ over {0.5, 1.0, 2.0}. Separately, they run Bloqade analog emulations of square atom arrays driven by a ramp–hold–ramp pulse reaching Ω_max = 5 rad/μs with total pulse area matching a flat π/2 pulse, defining a fidelity proxy F = 1 − (1/p_ideal)·(1/N)Σ|ρᵢ − p_ideal| with p_ideal = 0.5. The same experiment is run on real Aquila hardware for square arrays of side m ∈ {1,2,3,4} (N = m² atoms) with nearest-neighbor spacing swept over a ∈ {4,5,6,7,8,9,10,11} μm, 1000 shots each. Finally, a digital study uses the GeminiOneZoneNoiseModel to compute Hilbert–Schmidt fidelity for compiled circuits.

Why This Matters

Impact on research. The paper claims to be the first work to formally connect energy, runtime, and fidelity in Rydberg-atom quantum computers, supplying the systems-level abstraction that prior work on compilation, control, and algorithmic fidelity left out. It converts a device-physics parameter (the blockade ratio) into a scheduling knob, which reframes qubit packing as a first-class control variable rather than an implementation detail.

Real-world applications:

• Quantum cloud job scheduling: providers charging per-task could use fidelity-targeted packing to raise throughput per machine without violating user-specified accuracy.
• Variational and sampling workloads: algorithms requiring thousands of repeated shots of the same circuit are exactly the case where intra-device parallelism amortizes static power.
• Datacenter capacity and thermal planning: since power is dominated by a static baseline, operators can predict facility power needs for future qubit counts (roughly 10 kW at 10,000 qubits in the paper's projection) rather than assuming cryogenic-scale costs.
• Hardware co-design: the r_b/d₀ ratio tells designers where to trade lattice pitch, Rabi frequency, and detuning against achievable parallelism.

Industry relevance. The work targets commercially available hardware (QuEra Aquila via AWS Braket, and the Gemini digital system), and PaQit is explicitly pitched as software-only, requiring no hardware modifications and deployable inside an existing quantum control stack. That makes it immediately relevant to cloud quantum providers and to anyone budgeting quantum compute by energy rather than by wall-clock time.

Future Directions

• Replacing proxy fidelity with full process fidelity. The authors explicitly note their analog metric is a state-preparation fidelity proxy, not full quantum state or process fidelity; extending validation to gate-level and algorithm-level fidelity is an open step.
• Calibrating the normalized constants. The paper sets κ = 1.0 and f_max = 1.0 as normalized values; hardware-specific calibration of κ against pulse shape and detuning is left open.
• Integrating with real control stacks. PaQit is described as implementable as a scheduler in the control stack and recomputable when calibration changes, but an actual deployed scheduler and its overhead are not reported.
• Extending beyond the studied regimes. The sweeps cover r_b/d₀ ∈ {0.5, 1.0, 2.0}, fidelity 0.8–1.0, and digital circuits on at most 9 line qubits; whether the trends hold for larger digital workloads with all-to-all connectivity is not established.

Target Audience

Quantum-architecture and quantum-systems researchers, compiler and scheduler developers for quantum cloud platforms, and hardware engineers working on neutral-atom devices who need to reason about parallelism versus crosstalk. It also suits energy-aware computing researchers interested in a case study where static power dominates, and graduate students looking for a worked example of translating device physics into a scheduling policy. Readers wanting detailed algorithmic benchmarking or full process-tomography validation will not find it here.

Authors’ abstract

Neutral-atom quantum computers provide a scalable platform for large-scale quantum computation due to their all-optical control, room-temperature operation, and flexible lattice geometry. Although the favorable energy characteristics of these systems are well recognized, the relationship between system-level energy consumption, runtime, and computational fidelity remains poorly understood, limiting practical scheduling decisions. In this work, we develop a hardware-grounded analytical model that captures how energy and runtime scale with qubit utilization in neutral-atom systems. We introduce PaQit, a fidelity-aware qubit packing framework that integrates device-level Rydberg interaction physics with system-level scheduling to jointly optimize energy, runtime, and fidelity. By translating fidelity targets into packing decisions, PaQit identifies operating regimes that maximize parallelism while respecting interaction-driven crosstalk constraints. We validate the analytical framework using simulations of QuEra's analog Aquila system and digital Gemini system, as well as real hardware executions, demonstrating close agreement between the predicted trends and observed system behavior.