Reliable Linear, Sesquilinear and Bijective Operations On Integer Data Streams Via Numerical Entanglement

摘要: A new technique is proposed for fault-tolerant linear, sesquilinear and bijective (LSB) operations on M integer data streams (M ≥ 3), such as: scaling, additions/subtractions, inner or outer vector products, permutations convolutions. In the method, input are linearly superimposed to form numerically-entangled that stored in-place of original inputs. series LSB can then be performed directly using these entangled streams. The results extracted from output by additions arithmetic shifts. Any soft errors affecting any single disentangled stream guaranteed detectable via a specific post-computation reliability check. addition, when utilizing separate processor core each streams, approach recover all outputs after fail-stop failure. Importantly, unlike algorithm-based fault tolerance (ABFT) methods, number required entanglement, extraction validation related inputs does not depend complexity operations. We have validated our proposal in an Intel (Haswell architecture with AVX2 support) several types operations: fast Fourier transforms, circular convolutions, matrix multiplication Our analysis experiments reveal incurs between 0.03% 7% reduction processing throughput wide variety This overhead 5 1000 times smaller than equivalent ABFT method uses checksum stream. Thus, used faultgenerating hardware safety-critical applications, where high without cost modular redundancy.