Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Demonstration of a Logical Architecture Uniting Motion and In-Place Entanglement

Rich Rines, Benjamin Hall, Mariesa H. Teo, Joshua Viszlai, Daniel C. Cole, David Mason, Cameron Barker, Matt J. Bedalov, Matt Blakely, Tobias Bothwell, Caitlin Carnahan, Frederic T. Chong, Samuel Y. Eubanks, Brian Fields, Matthew Gillette, Palash Goiporia, Pranav Gokhale, Garrett T. Hickman, Marin Iliev, Eric B. Jones, Ryan A. Jones, Kevin W. Kuper, Stephanie Lee, Martin T. Lichtman, Kevin Loeffler, Nate Mackintosh, Farhad Majdeteimouri, Peter T. Mitchell, Thomas W. Noel, Ely Novakoski, Victory Omole, David Owusu-Antwi, Alexander G. Radnaev, Anthony Reiter, Mark Saffman, Bharath Thotakura, Teague Tomesh, Ilya Vinogradov

Abstract

We demonstrate a logical neutral atom architecture that integrates atom motion with in-place entanglement to achieve lower overheads than entangling-zone approaches. Using a 114-qubit device, we perform three proof-of-principle logical-qubit experiments. First, we implement a pre-compiled, non-scalable variant of Shor's algorithm, observing improved logical-over-physical performance, including with loss correction and leakage detection, achieving up to a 2x reduction in TVD. Second, we construct constant-depth logical CX ladders; on current hardware these execute with serial entangling operations, yet still yield 2-4x lower error for 8 and 12 logical qubits. Third, we prepare the [[16,4,4]] code and perform single-round decoding with post-processed error correction, achieving 8x improvement on logical vs physical. These results demonstrate how combining motion with in-place entanglement offers lower overhead than entangling-zone approaches.

Demonstration of a Logical Architecture Uniting Motion and In-Place Entanglement

Abstract

We demonstrate a logical neutral atom architecture that integrates atom motion with in-place entanglement to achieve lower overheads than entangling-zone approaches. Using a 114-qubit device, we perform three proof-of-principle logical-qubit experiments. First, we implement a pre-compiled, non-scalable variant of Shor's algorithm, observing improved logical-over-physical performance, including with loss correction and leakage detection, achieving up to a 2x reduction in TVD. Second, we construct constant-depth logical CX ladders; on current hardware these execute with serial entangling operations, yet still yield 2-4x lower error for 8 and 12 logical qubits. Third, we prepare the [[16,4,4]] code and perform single-round decoding with post-processed error correction, achieving 8x improvement on logical vs physical. These results demonstrate how combining motion with in-place entanglement offers lower overhead than entangling-zone approaches.

Paper Structure

This paper contains 35 sections, 13 equations, 13 figures, 1 table.

Table of Contents

  1. Introduction
  2. Methods
  3. Movement
  4. Logical Encoding
  5. Post-Processing
  6. Leakage Detection
  7. Noisy Simulations
  8. Shor's Algorithm (Pre-Compiled)
  9. Logical circuits
  10. Results
  11. Constant-Depth CX ladders
  12. To Constant Depth
  13. Results
  14. Many-Hypercube Code
  15. Logical Circuits
  16. ...and 20 more sections

Figures (13)

  • Figure 1: (a) Sqale's architecture, combining locally addressed gates and mid-circuit rearrangement. (b) Fluorescence images before (top) and after (bottom) translating and stretching a 6-qubit block through interstitial lanes. (c) Summary of demonstrations in this work. Upper panel: logical qubit layout and mid-circuit reconfigurations for (i, ii) [[4, 2, 2]]-encoded demonstration of Shor's algorithm and a constant-depth CNOT (CX) ladder, and (iii) state preparation in the [[16, 4, 4]] many-hypercube code. Lower panel: Encoded/unencoded performance in terms of total variation distance (TVD).
  • Figure 2: Experimental results for all Shor circuits run on hardware. (a) The postselection discard, and (b) The TVD computed between the experimental and ideal distributions in (a). The error bars represent a 68% confidence interval (CI) on the TVD for the narrowest CI obtained through minimization of the CI width. Diamond data points represent the expected (mean) TVD from a default Sqalesim noise sampling for the same number of experimental shots, with error bars on the mean of ten trials.
  • Figure 3: An "outside-in" implementation of a CX ladder consisting of 5 CXs. (a) An implementation on a $6\times1$ line (or $3\times2$ grid) of qubits with nearest-neighbor connectivity. (b) A rearrangement of (a) via row-major ordering. The CX ladder places the sum (modulo 2) of every individual qubit's state on the middle/last qubit for sub-figures (a)/(b) respectively. The dashed barriers delineate segments of the circuit that can be executed in parallel and the brackets denote qubits within the same row.
  • Figure 4: Logical circuit for a 5 CX, constant-depth CX ladder on a $4\times2$ grid of qubits. The brackets surround qubits on the same row. The dashed lines separate the circuit into three parts: the first creates a Bell pair between the ancilla row and the second data row, the second consists of the CXs from the first two sections of the original circuit (Figure \ref{['fig:cdcx']}) being executed in parallel, and the third consists of the final CX from the original circuit, measurement, and classical correction.
  • Figure 5: Error rates with 68% confidence intervals for unencoded (gray, wide) vs encoded (green, narrow) constant-depth CX ladders with 8 logical qubits.
  • ...and 8 more figures