SYSTEMS LAB

Research tracks with explicit evidence state.

Research tracks, smaller experiments, validation methods, and work in progress.

AMD XDNA1 / AIE2

VALIDATINGHIGH CONFIDENCE

Native NPU execution, kernel mapping, runtime behavior, and CPU/NPU correctness checks.

Evidence progress 92%Integrity 83% · GOOD

MLIR-AIE / IRONXRTTransformer kernelsCPU/NPU validation

Latest experimentEXP-2026-08-02-NPU-VALIDATION · Pinned native NPU validation baselinePASSED · 2026-08-02

Next milestoneExpand reproducible decoder-kernel validation and publish benchmark conditions alongside results.Last updated 2026-08-29

Open evidence-backed case study →

Firmware research

RESEARCHMEDIUM CONFIDENCE

Extracting, comparing, and documenting firmware behavior with reproducible artifacts.

Evidence progress 84%Integrity 67% · GOOD

Binary analysisFirmware imagesHardware validationRecovery-aware workflows

Latest experimentEXP-2026-08-27-PROVENANCE · Release provenance and repository-hardening passPARTIAL · 2026-08-27

Next milestoneAttach hashes, extraction artifacts, and recovery-aware reproduction steps to each hardware claim.Last updated 2026-08-29

Open evidence-backed case study →

NPU compute experiments

ACTIVEMEDIUM CONFIDENCE

Exploring what first-generation Ryzen AI NPUs can do outside conventional inference paths.

Evidence progress 84%Integrity 67% · RESEARCH

XDNA runtimeCompute kernelsMeasurementFailure analysis

Latest experimentEXP-2026-08-10-PREFLIGHT · Testnet preflight handoffPARTIAL · 2026-08-10

Next milestoneSeparate measured accelerator work from host orchestration and record repeatable failure modes.Last updated 2026-08-29

Open evidence-backed case study →

Research contract

Every claim should identify what was observed, what remains an interpretation, how it can be reproduced, which artifacts and commits support it, and which limitations still block a stronger conclusion.