Huliang Quantum Publishes Three DAC 2026 Papers: 1000x Faster CNOT Synthesis, 606x Faster Circuit Optimization, and 95% Error Reduction for Neutral-Atom QEC

Forum topic · 小凯 · 2026-08-15

Summary

Chinese quantum computing company Huliang Quantum (Arc Light Quantum) had three papers accepted at DAC 2026, covering quantum circuit compilation across synthesis, optimization, and hardware-aware scheduling. First, Lin-search performs exact optimal CNOT circuit synthesis using two-tier lower-bound pruning and canonical-form equivalence merging, achieving up to ~1000x speedup over Qiskit-SAT on small-to-medium circuits and reducing consecutive CNOT subcircuits by up to 26% in Clifford+T optimization flows. Second, a hypergraph-based circuit optimization method replaces DAG template matching by merging same-direction adjacent gates into supernodes, cutting runtime from ~3.5 hours to ~21 seconds (606x) on a 2,707-gate, 48-qubit arithmetic circuit with identical gate-count quality. Third, the NEAT compiler jointly optimizes data-atom placement, ancilla atom movement, and gate scheduling for neutral-atom quantum error correction via symbolic constraint solving, reducing surface-code movement rounds from 100–624 to 3–4 and lowering end-to-end logical error rates to 1/2–1/20 of the Enola compiler baseline (up to 95% reduction).

Design automation's top conference, DAC 2026, has concluded with three accepted papers from the Huliang Quantum team — optimal CNOT circuit synthesis (Lin-search), quantum circuit optimization (hypergraph method), and neutral-atom quantum error correction compilation (NEAT). Though the three works attack different problems, they share one goal: making quantum programs run better on real hardware with fewer resources and lower cost. All three achieve order-of-hundred efficiency improvements in the best cases.

Breakthrough 1: Lin-search — CNOT Synthesis Up to 1000x Faster

The CNOT gate is the most common two-qubit gate in quantum circuits and among the most error-prone operations — typically with error rates well above single-qubit gates. Reducing CNOT count directly improves circuit execution quality.

Existing compilers (e.g., Qiskit-SAT) can find "fairly short" circuits but provide no guarantee of optimality. The Huliang Quantum team pursued exact synthesis: given a target function, find the implementation with the minimum number of CNOT gates.

Returning to first-principles algorithms, they search directly in the space of CNOT circuits with a method called Lin-search. The core is two-tier lower-bound pruning:

"Distance-to-target rows" lower bound: back-propagating from output differences to determine the minimum number of additional gates needed
"Cut-rank" lower bound (independent information that must be transmitted): determining how much cross-region information must still pass through CNOTs

Combined with canonical forms that automatically merge equivalent circuits, Lin-search finds gate-optimal implementations on an ordinary laptop within a 600-second time limit.

Measured results: for a 4-qubit, 5-gate circuit, naive search must check 248,832 candidates; with the two lower bounds plus equivalence merging, only 18 candidates need checking, finding the optimum in 0.27 milliseconds. In randomized tests at 15 qubits and 15 CNOT gates, Lin-search extends the tractable scale 3 qubits beyond Qiskit-SAT, with hundreds-fold speedups on small-to-medium circuits, up to ~1000x.

In deployment: embedding Lin-search into a Clifford+T circuit optimization flow reduced consecutive CNOT subcircuits by 16% on average (up to 26%), with 30% fewer tasks timing out compared to Qiskit-SAT.

Breakthrough 2: Hypergraph Method — 606x Faster Optimization

Quantum programs are usually optimized before running on hardware, but commutation relationships between gates cause the search space to explode. Existing methods use DAGs (directed acyclic graphs) — preserving non-commuting dependencies and removing commuting edges — which turns commuting gates into isolated nodes. Algorithms get "lost" on large circuits and are forced into blind full-graph searches.

Huliang Quantum introduces a hypergraph representation:

Adjacent gates acting in the same direction on the same qubit are merged into "supernodes"
"Hyperedges" restore cross-qubit connections
The algorithm can then precisely expand matches outward from local neighborhoods

Compared head-to-head against Qiskit's DAG template matcher:

| Circuit Scale | Qiskit | Qiskit (heuristic) | Hypergraph method | Speedup | |---|---|---|---|---| | 100-gate random | 27.64s | 18.97s | 1.75s | 15–65x | | 300-gate random | 177.5s | 93.2s | 5.67s | 15–65x | | 500-gate random | 583.34s | 293.85s | 9.19s | 11–29x | | 2,707-gate / 48-qubit arithmetic | ~3.5 hours | — | ~21 seconds | 606x |

3.5 hours vs. 21 seconds — a 606x speedup — and post-optimization gate counts match Qiskit, with no quality trade-off. The advantage grows with circuit size, showing potential for large-scale quantum circuit processing.

Breakthrough 3: NEAT Compiler — 95% Error Reduction for Neutral-Atom QEC

Neutral-atom platforms are a hot route in quantum hardware today (QuEra, Atom Computing, Pasqal, Vector Quantum, etc.). Their quantum error correction (QEC) relies on ancilla atoms shuttling through the array to extract error syndromes — but past compilers optimized "gate execution order" and "atom movement paths" separately, like designing bus timetables and driving routes independently, resulting in massive redundant movement time and increased error.

Huliang Quantum's NEAT compiler puts three variables into a single symbolic constraint model:

Where data atoms sit (fixed coordinates)
How ancilla atoms move (position at each time step)
Which parallel time step each quantum gate executes in

The solver simultaneously satisfies logical constraints (QEC code check matrices) and hardware constraints (Rydberg interaction range, laser timing, collision-freedom, relative ordering of parallel trajectories). The team designed a two-tier symmetry-breaking scheme — lightweight constraints first fix temporal direction and spatial orientation, then lexicographic constraints handle hard instances — preventing the solver from repeatedly exploring equivalent solutions.

Measured results (baseline: the Enola compiler):

| QEC Code | Enola Move Rounds | NEAT Move Rounds | Enola Total Distance | NEAT Total Distance | |---|---|---|---|---| | Surface7 | 177 | 3 | 3907.2 | 568.6 | | Surface13 | 624 | 4 | — | — | | Planar7 | 100 | 3 | 1291.6 | 355.8 | | Planar13 | 292 | 3 | 11115.3 | 1580.1 |

End-to-end logical error rate: under identical noise conditions, NEAT-compiled circuits achieve error rates 1/2 to 1/20 of Enola's — a reduction of up to 95%.

Three Takeaways

First, compilation engineering is the real battleground. Quantum hardware keeps scaling, but whether "algorithms can actually run on real hardware" is decided by compilers. Huliang Quantum's three papers cover three stages — synthesis (finding optimal gate counts), optimization (local acceleration), and mapping/scheduling (hardware awareness) — with order-of-hundred improvements at each link in the toolchain.

Second, the neutral-atom route is being accelerated by engineering. NEAT compresses surface-code correction movement rounds from 100–600 down to 3–4, cutting total distance by an order of magnitude. This means neutral-atom companies like QuEra, Atom Computing, and Vector Quantum can enter the "engineering-scale error correction" stage earlier, without waiting for qubit counts to multiply several times over.

Third, DAC is the true main battleground for quantum compilation. DAC (Design Automation Conference) is the premier venue for traditional EDA design and the primary arena for quantum design automation (QDA). All authors of Huliang Quantum's three papers come from its team and collaborator groups at the Chinese Academy of Sciences, Tsinghua University, and East China Normal University — evidence that stable output capacity in quantum compilation engineering has formed in China.

Over the next 6–12 months, watch two signals: 1) whether these three works get integrated into mainstream SDKs like Qiskit, Cirq, PennyLane, and Tket; 2) whether neutral-atom vendors (QuEra, Atom Computing, Vector Quantum, Qudoor) cite NEAT data on public benchmarks.

Sources: Tencent News, Aug 14, Huliang Quantum DAC coverage, Photon Box QUANTUMCHINA, Aug 14, DAC 2026 program.