A Streaming Architecture for Quantum Error Syndrome Compression at 4 Kelvin

Scaling quantum computers with more qubits increases the data volume transferred through cables between the millikelvin (mK) and room-temperature (300 K) domains. Advances in cryogenic qubit control aim to nearly eliminate 300 K-to-mK (downstream) communication, leaving mK-to-300 K (upstream) communication as a bandwidth bottleneck (periodic qubit measurement and data transfer for decoding are still needed). One potential solution, assuming digital qubit readout, is an intermediate processing stage at 4 K using digital superconducting electronics, to bring all or part of the error decoding closer to the measurement target. However, resource and thermal constraints at 4 K limit the amount of processing possible, which degrades decoding accuracy. We propose IcePack, a streaming superconducting architecture for lossless quantum error syndrome compression. IcePack reduces the number of syndrome indices transmitted through spatial and temporal clustering, encodes the remaining indices more efficiently than traditional binary schemes, and microarchitecturally exploits superconducting delay elements and deep pipelines to satisfy stringent integration and thermal constraints. Our results demonstrate a 300× reduction in upstream data volume over digital readout without compression and $2.4-4 \times$ over emerging compression techniques. This contributes to a Pareto improvement, cutting per-qubit upstream thermal load by 11× and latency by 10×, all while being compatible with existing decoders and available superconductor electronics processes. Prototypes of the key system components were fabricated and experimentally validated.