IRIS

Distributed quantum computers connect smaller quantum chips through photonic interconnects. Non-local CNOTs then require state teleportation (RELOCATE) or gate teleportation (Re-CNOT), making inter-chip operations slower and more error-prone than local gates.

IRIS targets low-level instruction scheduling and is orthogonal to the initial mapping policy. The evaluation uses modular dual-species neutral-atom processors with separate compute and communication qubits, where the available communication qubits determine each chip’s EPR capacity.

1. Utility-Driven Lookahead with Multi-Candidate Block Scheduling (UMS)

   Look ahead only to future blocks that share qubits with the current block, while retaining the top-w candidate schedules to avoid early commitment.

2. EPR-Capacity-Aware Early Scheduling (EES)

   Move dependency-resolved future CNOTs and their teleportations across block boundaries when doing so adds no relocation and preserves available EPR capacity.

Relative to QuComm, IRIS reduces effective teleportation count by 34% on average (up to 65%) and improves program-schedule latency by 2.0× on average (up to 2.9×).

These are schedule-level results, not compilation speedups. Larger programs take longer to compile; the paper evaluates that trade-off over repeated shots. The QEC-cycle comparison uses a 2×2 DQC.

• 
  Won Joon Yun Ph.D. Student · The University of Texas at Austin
• 
  Dhilan Nag The University of Texas at Austin
• 
  Sneha Ballabh The University of Texas at Austin
• 
  Jiapeng Zhao Cisco Quantum Lab
• 
  Eneet Kaur Cisco Quantum Lab
• 
  Prof. Poulami Das ECE Professor · The University of Texas at Austin