Yale University researchers have developed Prism, a quantum compiler that reduces qubit routing errors by up to 30.8% across a range of workloads. The work addresses a key challenge in quantum computing: translating logical gates into physical instructions while minimizing the need for SWAPs, operations that introduce errors and slow down computation.
Prism achieves this by delaying commitment to a specific physical circuit for a gate until routing finds a suitable layout, a strategy that reduces routing SWAPs and physical depth. Across 25 workloads with 7 to 25 logical qubits, the compiler also lowers photon loss on device models, requiring 44.3% fewer attempts per retained shot at 16 logical qubits.
Prism Compiler Reduces Qubit Routing Errors Up To 31%
Across three distinct device graphs, the Prism compiler reduces qubit routing errors by 20.4 to 30.8 percent, as measured by a reduction in routing SWAPs relative to a conventional greedy search algorithm. The compiler’s design allows it to explore multiple potential implementations of a logical gate simultaneously, selecting the most efficient one only after assessing the available qubit connections.
For dual-rail cavity qubits, a specific type of quantum system, the logical gate benefits from Prism’s flexibility, possessing four loss-admissible realizations that the compiler keeps open during routing. This means Prism considers all four possible ways to implement the gate before committing to one, allowing it to choose the option that minimizes the need for SWAPs, exchanges of qubit states, which introduce errors.
The compiler formalizes which physical circuits implement a gate while respecting the fault model, and defines when a gate can execute under a given layout, ensuring that only valid and efficient circuits are used. Evaluations across 25 workloads, ranging from 7 to 25 logical qubits, demonstrate consistent performance gains.
Beyond reducing SWAPs, Prism also lowers physical depth, the time it takes to complete a quantum computation, by 8.8 to 10.1 percent, enabling faster processing. “Prism carries each logical gate across the logical–physical boundary as a set of realizations and commits to one when routing reaches a layout in which that realization can execute,” explain Evan McKinney, Kun Liu, Harshvardhan K. Babla, Yongshan Ding and Robert J. Schoelkopf, all affiliated with Yale University.
Dual-Rail Qubit Encoding Defines Logical and Erasure Spaces
Dual-rail qubit encoding establishes a defined space for both logical computation and error detection, with the logical state of a qubit encoded across two physical modes, creating a complementary erasure space. This encoding uses an isometry that maps logical basis states to the vacuum when photon loss occurs, a phenomenon demonstrably observed in dual-rail cavity, transmons and neutral atom systems.
The system fixes each data wire at a specific location and maps interactions to supporting device edges, defining an availability relation important for gate execution. The logical availability graph, a key component of this framework, determines which logical qubits can interact given a current physical layout.
The resulting logical availability graph for this gate is precisely the quotient of the occupied device graph by these rail pairs, a relationship that streamlines the optimization process. Visualizations within the work demonstrate how different layouts impact the logical availability graph and necessitate physical SWAPs to achieve connectivity. Logical rotations, such as and , require adjacent rails for certain operations, while intermediate erasure checks also depend on this proximity. However, the system allows rails to separate temporarily, enabling a wider range of realization choices during routing, provided the logical Z value is preserved.
Prism Delays Gate Realization Commitment Until Routing Completion
By delaying commitment, Prism avoids forcing unnecessary qubit swaps that would otherwise increase the complexity and duration of quantum computations. The compiler assigns a routing cost to each unresolved gate, calculated as the minimum distance on the device graph between the qubits involved in a given realization, enabling informed decisions during the routing process.
“Each LoweredOp remains unresolved until the current layout λ supports a realization,” the paper explains, detailing how Prism dynamically adapts to the available physical connections. This dynamic approach contrasts with Logical-SABRE, which routes on a fixed logical coupling graph before lowering, and highlights Prism’s ability to optimize for the specific constraints of the physical hardware.
Logical CR_z Realizations Map to Device Graph Quotients
This approach allows for a subtle evaluation of gate execution possibilities, moving beyond simple binary availability. The compiler uses this multiplicity, maintaining all four options until the routing process determines the most efficient path. This strategy is formalized through the concept of a logical availability graph, which, for these qubits, is defined as the quotient of the device graph by the rail pairs.
The compiler’s definitions of admissible realization circuits and interaction patterns are not limited to the current encoding, offering potential for broader application. While some circuits may require routing even between their own interactions, the framework allows for the creation of a logical graph weighted by routing cost, further refining optimization. The work highlights that Prism retains multiple realizations not to limit compilable programs, but to ensure a single-interaction realization of any logical gate can be made executable on any connected device graph.
Prism’s Boundary IR Exposes Physical Wires for Optimization
Unlike conventional methods, Prism maintains multiple potential implementations of each logical gate as it crosses the boundary between logical and physical qubits, allowing the router to select the most efficient option based on the available connections. This approach fundamentally reframes code expansion and routing as a phase-ordering problem, similar to those encountered in quantum optimization passes. The compiler’s boundary intermediate representation (IR) exposes the physical wires connecting qubits while preserving the set of alternative circuits for each gate, enabling a more informed routing process.
During routing, Prism evaluates these alternatives under the current layout, using their associated costs to guide the insertion of SWAPs, operations that exchange qubit states and minimize overall circuit complexity. A logical qubit, defined as occupying the single-excitation subspace of two cavity modes or “rails,” is particularly sensitive to photon loss, a detectable error state that Prism’s approach aims to mitigate.
Prism Reduces SWAPs and Depth Across Multiple Workloads
On grid and hexagonal lattices, reductions in routing SWAPs persisted even as simulations scaled to 32 logical qubits, demonstrating the compiler’s effectiveness with increasingly complex quantum systems. The work directly addresses photon loss, a significant impediment to reliable quantum computation, by reducing the number of attempts needed to obtain a valid measurement result. Specifically, Prism requires 44.3% fewer attempts per retained shot at 16 logical qubits, as a geometric mean over workloads and device graphs.
This improvement stems from minimizing the number of locations where photons can be lost during qubit routing. Prism’s approach to managing dual-rail cavity qubits is central to its success; the compiler maintains four possible implementations of the logical gate, one for each rail pairing, until routing determines the optimal configuration. This delayed commitment allows the system to select the realization with the fewest potential loss locations, a strategy that demonstrably lowers decoded logical error in a three-round Steane memory when accounting for skip-gate leakage.
Prism Lowers Decoded Logical Error in Steane Memory
The resulting shorter routes also translate to a 44.3% reduction in the number of attempts needed to obtain a valid measurement result. The compiler maintains these multiple realizations throughout the lowering process, allowing it to select the most efficient implementation based on the final routed layout. This approach formalizes the distinction between logically correct, admissible and retained realization sets, ultimately impacting the fidelity of quantum computations.
“Routing changes which realization of an encoded gate can execute without SWAPs, so Prism keeps each gate’s realizations open until the router can execute one,” explains the paper’s methodology. Evaluations on modeled dual-rail cavity hardware demonstrate a significant reduction in decoded logical error within a three-round Steane memory, even under skip-gate leakage.
👉 More information
🗞 PRISM: A Logical-Physical Boundary Compiler for Dual-Rail Qubits
✍️ Evan McKinney, Kun Liu, Harshvardhan K. Babla, Robert J. Schoelkopf and Yongshan Ding
🧠 ArXiv: https://arxiv.org/abs/2611.23009v1




See today’s quantum computing news on Quantum Zeitgeist for the latest breakthroughs in qubits, hardware, algorithms, and industry deals.
