THE QUANTUM INTELLIGENCER

Two technical papers command the engineer's attention this morning. The first, on the Hamming Quasi-Cyclic key encapsulation scheme, presents a unified encoder datapath that serves all three security levels at runtime, reducing aggregate resource consumption by as much as 2.9 times when compared to a trio of fixed-level cores. The second, addressing soft-output decoding for cluster-based quantum decoders, introduces bounded and extra-cluster gaps that lower computational cost and require no alteration to existing FPGA implementations. The HQC scheme, which the National Institute of Standards and Technology has selected for standardization as a key encapsulation mechanism, yields per-level figures beyond the aggregate saving. At HQC-128 the unified datapath is over-provisioned, its area-delay product matching the best prior standalone encoder; at HQC-256 it improves upon that encoder by a factor of 1.65, and by as much as 2.55 times over single-level encoder components taken from complete KEM implementations. These are post-synthesis measurements on field-programmable gate arrays, reported by the authors. The decoding paper grounds its improvements on the observation that the precise value of a large soft output is often unnecessary. Its bounded cluster gap terminates the calculation early, improving the scaling with code distance over the 2024 method of Meister and colleagues; its extra-cluster gap requires only a small additional growth of the clusters produced by the decoder, so that soft outputs may be computed without altering the existing architecture of FPGA-implemented Union-Find decoders. The source documents supply further particulars. The decoding paper identifies the weakness of the earlier method of Meister and colleagues: in parallel environments its computational complexity rivals that of the Union-Find decoder itself, and it demands global information about the decoding graph, which existing FPGA implementations do not provide. The present work's two soft-outputs address exactly those deficiencies, the bounded cluster gap terminating calculation early and the extra-cluster gap requiring only a small additional growth of the decoder's clusters. On the HQC side, the encoder unites a configurable Reed-Solomon encoder with a single-cycle Reed-Muller stage and a direct stage interface that dispenses with intermediate buffering. Both papers are preprints; their claims await independent verification. The reader should mark the bounds of these claims. The HQC results concern the encoder alone, not the full key encapsulation mechanism, and they are measured by area-delay product on field-programmable gate arrays. Whether the same unified datapath serves as well on application-specific circuits, or within a complete KEM implementation, the paper does not say; the reported improvements of 1.65 and 2.55 times apply to the encoder component in isolation. The decoder work is narrower still. It addresses cluster-based decoders of the Union-Find family, and it addresses only the evaluation of soft outputs, not the decoding itself. The two gaps are offered as cheaper substitutes for the precise soft-output value, and the extra-cluster gap requires no alteration to existing FPGA-held Union-Find architecture; but the technique does not extend to other decoder paradigms, nor does it improve the hard-decision decoding proper. Each result is a step along a particular corridor, not a general road. —Ada H. Pemberley Dispatch from The Prepared E0