Quantum Computing Report

QuEra, Harvard, and UMD Researchers Develop Polynomial-Time Classical Benchmarking for Non-Clifford Fault-Tolerant Circuits

Researchers from QuEra Computing, Harvard University, the Joint Center for Quantum Information and Computer Science (QuICS) at the University of Maryland, UC Berkeley, and ETH Zürich have introduced a polynomial-time Monte Carlo framework for benchmarking the logical fidelity and Pauli observables of fault-tolerant non-Clifford quantum circuits. While non-Clifford gates such as T and controlled-controlled-Z (CCZ) are required for universal quantum computation, they typically prevent efficient stabilizer simulation by propagating Pauli errors into non-Pauli operators.

Affine–Diagonal Benchmarking Framework & Tricycle Factory Comparison
Component / BenchmarkTechnical Scope & Noise ModelKey Scaling & Performance Findings
Affine–Diagonal SimulatorCircuits with X, CNOT, and 3rd-level diagonal gates (T, CS, CCZ), arbitrary Clifford layers, and teleportation feedforward.Exact syndrome sampling and logical fidelity in O(n3T) time per sample, independent of logical qubit count k; extends check expectation values to 4th-level diagonal gates.
Hypercube-IQP ScalingEncoded [[8, 3, 2]] hypercube circuits scaled up to n = 256 physical qubits on a single CPU.Confirmed polynomial scaling (n2.51 precomputation) with per-sample cost decoupled from logical block rate.
[[27, 3, 3]] Tricycle Factory (Protocols A–E)Three 27-qubit blocks (81 data qubits) executing depth-2 transversal CCZ at p = 10−3 and pCCZ = 2 × 10−3 (10:1 Z-bias).Single shared Steane ancilla block (Protocol E: 81 CNOTs, 3 layers) achieved the highest acceptance-to-fidelity efficiency, outperforming per-block Steane (Protocol D) and bare-ancilla extraction (Protocol C).

To bypass the exponential cost of full state-vector or magic-state simulation, the authors observed that evaluating logical output fidelity does not require simulating logical measurement outcomes. By backpropagating circuit-level stochastic Pauli faults, stabilizer checks, and Pauli recovery operations through affine-diagonal circuits to the input, propagated errors reduce to Clifford operators whose pairwise commutators are diagonal Pauli strings, enabling exact syndrome sampling via binary Gaussian elimination and polynomial-time stabilizer overlap calculations.

The team demonstrated the method on hypercube instantaneous quantum polynomial (IQP) circuits up to 256 physical qubits, magic-state cultivation routines, and high-rate magic-state factories built from three copies of a [[27, 3, 3]] tricycle code. Comparing five post-CCZ Z-syndrome extraction strategies, the simulations showed that using a single shared ancilla block to check all three data blocks simultaneously recovers circuit-level fault protection with lower qubit and CNOT overhead than dedicated per-block ancilla extraction.

Read the full research paper on arXiv here.

November 20, 2026

Exit mobile version