Skip to content

SC-NeuroCore JOSS Pre-Submission Draft

Miroslav Šotek · ORCID 0009-0009-3560-0851 · Anulum Research, Independent Researcher

Date: 14 May 2026

Status: Pre-submission draft. Formal JOSS submission is postponed until the remaining production-hardening, validation, coverage, documentation, and hardware-evidence TODOs are complete.

Software DOI: 10.5281/zenodo.18906614

The canonical JOSS source is paper/paper.md at the repository root. This page mirrors the current submission stance for the public documentation site: the paper is being kept in JOSS format, but it is not yet being submitted.

Current Evidence Boundary

SC-NeuroCore is an open-source Python and Rust framework for stochastic- computing experiments with spiking neural networks. It provides deterministic bitstream simulation, fixed-point export utilities, hardware-oriented intermediate representations, NIR bridge work, Rust acceleration paths, and Verilog-generation collateral.

Public claims are limited to committed artefacts and CI-verifiable behaviour. The current package version is 3.14.0, and the current Python coverage gate is 96%. The project target remains 100% coverage, but 100% is not a current release claim.

Submission Blockers

The JOSS submission/review item remains open until:

  • numeric claims in the paper and public docs are regenerated from current CI, release, benchmark, and artefact evidence;
  • physical FPGA claims are backed by PYNQ-Z2 board logs for power, thermal, latency, and parity;
  • unresolved model-fidelity and documentation audit items are closed or explicitly scoped;
  • core installation, examples, tests, contribution paths, and support channels have been checked from a clean reviewer-style environment.

JOSS Structure

The draft follows the current JOSS paper structure:

  • Summary
  • Statement of Need
  • State of the Field
  • Software Design
  • Research Impact Statement
  • AI Usage Disclosure
  • Acknowledgements
  • References

Research Impact Position

The near-term research value is a reproducible experimental surface for stochastic-computing SNN hardware studies: bitstream encoders, training/export utilities, NIR adapters, Rust kernels, HDL generators, formal checks, benchmark scripts, and documentation live in one inspectable repository. Active collaboration work around SHD deployment evidence and stochastic SNN compression remains pre-submission evidence until the validation queue is complete.