arXiv:2610.17058 · IEEE International Symposium on Workload Characterization
Architecting Instruction Execution for Scalable Fault-Tolerant Quantum Computing
Dmitry Filippov, Justin Hogaboam, Mathias Soeken, Raghu Gatta, Prakash Murali, Matthias Troyer
Abstract
Scaling fault-tolerant quantum computing (FTQC) requires orchestrating millions of operations continuously across a tightly coupled classical-quantum control stack. A critical challenge in this orchestration is determining the optimal hardware-software boundary, i.e., the Quantum Instruction Set Architecture (QISA). We focus on cryogenic, fixed-connectivity qubit platforms, such as superconducting and topological Majorana qubits. For these, existing QISA choices either operate at too low a level, violating cryogenic thermal budgets at scale, or at too high a level, imposing unrealizable complexity on the control engine. To reason about these tradeoffs, we introduce the ARIES instruction set hierarchy, which categorizes QISAs into distinct abstraction levels. To systematically evaluate these levels, we develop an end-to-end evaluation toolflow featuring a cycle-accurate classical driver and hardware engine simulator. Using eight quantum applications across superconducting and topological qubit platforms, we measure instruction bandwidth, transceiver power dissipation, and measurement processing overheads. We demonstrate that while physical-level abstractions exceed the thermal limits of dilution refrigerators by orders of magnitude, canonical-level abstractions significantly underestimate hardware routing congestion and execution runtimes. Our Level 4 QISA (local logical operations) offers an optimal architectural balance. As quantum hardware platforms scale toward FTQC in the coming years, our findings provide concrete architectural guidance for the design of their classical control stacks and instruction interfaces.
AI Summary
The paper asks where to place the software–hardware instruction boundary for fault-tolerant quantum computers with cryogenic control and fixed two-dimensional connectivity. It introduces ARIES, a five-level QISA hierarchy spanning physical gates, QEC micro-operations, QEC-native logical primitives, local logical operations, and hardware-agnostic canonical operations. An evaluation framework combines Q# compilation, a Rust cycle-accurate driver/engine simulator, analytical bandwidth and transceiver-power models, and FPGA synthesis/resource modeling. Eight application benchmarks cover condensed-matter simulation, factoring kernels, and chemistry walk operators, using superconducting surface-code and prospective Majorana surface-code/Hastings–Haah configurations. Power estimates assume $0.5$ pJ/bit transceivers and a $1.5$ W cooling budget at 4 K. Physical-level instruction streams require 55 Tbps–20 Pbps without SIMD and 700 Gbps–62 Tbps with SIMD. Moving from QEC micro-operations to QEC-native logical instructions, with implicit background error correction, reduces modeled power by an average $263\times$. The recommended local-logical interface further reduces issue bandwidth by $1.5\times$, return bandwidth by $2.6\times$ (up to $5.2\times$), and power by $1.6\times$, while accelerating factoring and chemistry kernels by roughly $1.2$–$1.4\times$. Resource modeling estimates 40,000 versus 260,000 LUTs at 4,096 logical qubits for local-logical versus canonical engines. Canonical-level timing assumptions hide routing costs; realistic simulated cycle counts average $7.3\times$ the perfect-driver baseline. The stated scope excludes trapped-ion and neutral-atom architectures. Larger algorithms are represented by sampled kernels rather than complete executions. Physical-level bandwidth is calculated analytically, measurement outcomes are randomized, and compilation or parallelism-extraction improvements could reduce simulated runtimes.
Central claim may not hold: The claim that Level 4 offers an optimal architectural balance is stronger than the demonstrated recommendation for the particular implementations, workloads, and hardware assumptions evaluated. The Level 5 bandwidth comparison combines a different instruction interface with idealized remote-operation timing and connectivity, so its bandwidth penalty does not independently establish the superiority of Level 4. The reported low-end Level 1 SIMD bandwidths imply only 0.535 W of transceiver power at the stated energy per bit, contradicting the blanket claim that SIMD physical instruction interfaces exceed the 1.5 W cooling budget. The Level 4 feasibility argument does not explicitly account for QEC-decoder resources and power, although moving syndrome processing into the engine makes those costs relevant to the claimed cryogenic feasibility.
Written by gpt-6.1-sol from the paper's full text. Automated: it can be wrong.
- DOI
- 10.1109/IISWC71020.2026.00052
- Comments
- Published in IISWC 2026