Helios

Raveh Neeman

doi:10.5281/zenodo.22674814

Quantinuum · trapped ions · DEPLOYED (Nov 2025 (peer-reviewed Jun 2026)) · 98 physical qubits · access: cloud + on-premise + NVIDIA GB200 integration

architecture: Trapped ions — QCCD (transport between zones) · Open on the map

Register profile — the machine's variant of its architecture

qubit type: hyperfine clock states of Ba-137, chosen because its transitions are visible-wavelength

gate mechanism: laser Molmer-Sorensen RZZ(pi/2); arbitrary-angle ZZ; 8 high-fidelity gate zones, up to 8 two-qubit gates in parallel

connectivity: QCCD on a ~15 mm surface-electrode chip with a four-way X-junction and a ring-shaped rotating storage loop; all-to-all by transport

control: room-temperature electronics; 1,228 trap electrodes driven by 273 independent voltages (~2.8 control signals per qubit); shared time-multiplexed lasers, >=5 wavelengths (515 nm gates; 493, 614, 650, 1762 nm) delivered as focused free-space beams propagating parallel to the chip surface

control placement: room-temp + free-space optics

readout: fluorescence, destructive per ion; mid-circuit measurement and reset with microsecond-latency conditional branching

gate evidence: a number stated by the vendor, protocol unstated

decoder mode: offline — syndromes decoded after the run

codes: iceberg, many_hypercube, stab_16_4_4, stab_4_2_2, steane, stellated_dodecahedron_css

flags: code id missing in the register, conflicting claims, estimated baseline

Technologies by layer — the machine's cell per layer

1 Qubit carrier	Trapped atomic ion · primary ▣ Fig. 1 (98-ion image in 2-D trap); §Ion species: qubit and coolant ✅ 98 ¹³⁷Ba⁺ qubit ions (¹⁷¹Yb⁺ co-trapped as coolant); Ba chosen for visible-wavelength transitions (493/614/650/1762 nm), removing Yb's UV laser stack
2 Encoding	Hyperfine / clock-state qubit · primary ▤ §Ion species: qubit and coolant ✅ hyperfine clock states \|F=1,m_F=0⟩ / \|F=2,m_F=0⟩ of ¹³⁷Ba⁺
3 Gate mechanism	Mølmer–Sørensen / light-shift laser gate · primary ▣ Fig. 2d,e (4-ion crystal alignment, 2Q gate beams); Methods: Quantum operations ✅ laser MS RZZ(π/2), arbitrary-angle ZZ; 2Q infidelity 7.9(2)×10⁻⁴ across all pairs, 1Q 2.5×10⁻⁵; t_2q ≈ 70 µs; 8 gate zones in parallel (Jun 2026)
4 Connectivity / transport	Ion shuttling (QCCD, junctions, grid traps) · primary ▣ Fig. 2a–c (ring storage, cache, junction transport); Fig. 3 (time per layer) ✅ QCCD on ~15 mm 2-D surface-electrode chip, four-way X-junction + rotating ring storage loop; all-to-all by transport, degree 97; ~55 ms per circuit layer
5 Control	Free-space laser addressing of ions (AOM/AOD beams) · primary ▤ § Optical delivery ✅ focused free-space laser beams propagating parallel to the chip surface; ≥7 wavelengths across 8 zones Laser control with integrated photonics (ions) · alternate ▦ 🔎 integrated photonics: direction (Quantinuum roadmap), not used on Helios
6 Readout	Fluorescence state detection (ions, defects) · primary ▣ Fig. 4a,b,i (SPAM and MCMR component errors) ✅ fluorescence, destructive per ion; SPAM infidelity 3.3(5)×10⁻⁴ (Nature; secondary 4.8×10⁻⁴); mid-circuit measure + reset with µs-latency branching
7 Code	High-rate concatenated codes with transversal gates · primary ▤ ✅ [[20,2,6]] C₄-Helix (Sep 2026); iceberg [[2m,2m−2,2]] concatenated to d=4: 94 error-detected / 48 error-corrected logical from 98 physical Colour code (transversal Cliffords) · alternate ▤ 🔎 no colour-code result named for Helios in the note; codes run are iceberg/[[4,2,2]] concatenations and C₄-Helix Error-detection codes ([[4,2,2]], d=2 surface, iceberg, spacetime) · alternate ▤ Abstract ✅ iceberg [[k+2,k,2]] detection codes and two-level concatenated iceberg [[(k2+2)(k1+2),k2k1,4]] on 98-qubit Helios: 48–94 logical qubits beyond break-even, with post-selection Magic-state factory (cultivation / distillation / code switching) · alternate ▤ 🔎 no magic-state result named for Helios in the note; logical gates cycle-benchmarked at ~10⁻⁴ (Mar 2026)
8 Decoder	GPU decoding via NVQLink · primary ▤ 🔎 GPU real-time QEC via CUDA-Q QEC over NVQLink: BP-OSD median 67 µs round trip, 5.4× LER gain; record results decoded offline with Frontier MWPM / Sparse Blossom (+ correlated matching) · alternate ▤ 🔎 alternate; record decoding uses Frontier (offline), real-time uses GPU BP-OSD; no matching result named in the note
9 Interconnect	— none — nothing in this layer ◌ 🔎 single module — no ion–photon link on Helios; ic_ionphoton is Quantinuum's stated future path but nothing is built
10 Manufacturing	Ion-trap fabrication (surface-electrode chips; machined 3D blade traps) · primary ▤ §QPU architecture and ion trap design 🔎 in-house surface-electrode microfabrication; ~15 mm 2-D chip with 1,228 electrodes; cryogenic surface trap Optical / mechanical assembly (lasers, vacuum, objectives) · alternate ◌ 🔎 room-temperature optical bench; 515 nm gate light and 493/614/650/1762 nm SPAM light delivered through the trap; 1762 nm cavity-locked

Standard records

connectivity degree → 97 neighbours · best (all-to-all by transport)

Nov 2025 · source

2Q error under full-width parallel operation → 7.90e-4 · typical across all pairs with 8 zones active

Jun 2026 · source

control lines per qubit → 2.79 lines/qubit · typical (273 voltages / 98 qubits)

1,228 trap electrodes · Nov 2025 · source

SPAM error → 3.30e-4 · best

a secondary analysis gives 4.8e-4 · Jun 2026 · source

decode latency per round → 67 µs · typical (BP-OSD median on a GH200 over NVQLink)

5.4x LER gain · 2026 · source

transport per gate layer → 55 ms · typical (average time per circuit layer)

~5e4x a superconducting 1.1 us cycle · Jun 2026 · source

evidence: ✅ 8 / 16

Pictures

Every picture carries the story of what it shows and its credit line.

A fluorescence camera image of 98 trapped barium-137 ions lit by resonant laser light, with thin outlines drawn over it: a square 'Ring Storage' holding a ring of bright dots (each dot is one ion), a 'Cache' where two lines cross, and a long 'Quantum Logic' region. It is Fig. 1 of an arXiv paper by A. Ransford et al. (Quantinuum), dated 2025-11-07, showing the Helios 2D surface trap. Notice how memory (ring and cache) is kept apart from the processor zone; ions are shuttled between them through a junction. “Fig. 1 — 98 atomic ions illuminated by resonant laser light in the Helios 2D surface trap” by Ransford, A. et al. (Quantinuum), arxiv.org/abs/2511.05465, CC BY 4.0 (creativecommons.org/licenses/by/4.0/)

A multi-panel schematic drawing: (a) rings of coloured ion dots showing loading in batches, (b) the trap layout, a ring-shaped storage region joined through an X-shaped junction to a cache, a quantum-logic region and leg storage, (c) close-ups of ions crossing the junction, (d) an aligned ion pair, (e) crossed laser-beam layouts. It is Fig. 2 of an arXiv paper by A. Ransford et al. (Quantinuum), dated 2025-11-07, on the Helios design and operations. Follow the ions from storage through the junction to the lasers. “Fig. 2 — Illustration of the Helios design and conception of operations” by Ransford, A. et al. (Quantinuum), arxiv.org/abs/2511.05465, CC BY 4.0 (creativecommons.org/licenses/by/4.0/)

A composite data figure, not hardware: (a) average survival probability and (b) leakage population against sequence length for many qubit pairs, and (c) a bar chart of error (x10^-3) per qubit pair, split into computational and leakage parts. It is Fig. 6 of an arXiv paper by A. Ransford et al. (Quantinuum), dated 2025-11-07, on two-qubit randomized benchmarking in the Helios gate zones. Random gate sequences of growing length are run, and the decay rate gives the average error per gate; compare the leakage bars with the computational bars. “Fig. 6 — Two-qubit randomized benchmarking: survival probability, leakage and error by zone” by Ransford, A. et al. (Quantinuum), arxiv.org/abs/2511.05465, CC BY 4.0 (creativecommons.org/licenses/by/4.0/)