© 1998–2026 Miroslav Šotek. All rights reserved.
Contact: www.anulum.li | protoscience@anulum.li
ORCID: https://orcid.org/0009-0009-3560-0851
License: GNU AFFERO GENERAL PUBLIC LICENSE v3
Commercial Licensing: Available
# SC-NeuroCore
<p align="center">
<img src="docs/assets/sc_neurocore_header.png" width="1280" alt="SC-NeuroCore — Stochastic Computing & Neuromorphic Engine">
</p>
[](https://github.com/anulum/sc-neurocore/actions/workflows/ci.yml)
[](https://github.com/anulum/sc-neurocore/releases)
[](https://github.com/anulum/sc-neurocore)
[](https://anulum.github.io/sc-neurocore/)
[](https://www.gnu.org/licenses/agpl-3.0)
[](https://www.python.org/downloads/)
[](https://www.rust-lang.org/)
[](https://www.bestpractices.dev/projects/12175)
[](https://scorecard.dev/viewer/?uri=github.com/anulum/sc-neurocore)
[](https://reuse.software/)
[](https://colab.research.google.com/github/anulum/sc-neurocore/blob/main/notebooks/quickstart_colab.ipynb)
**Version:** 3.14.0
**Status:** 116 Neuron Models (107 Bio + 9 AI) | 99.49% MNIST (ConvSNN) | 2 928 Python tests passing (3 132 collected) + 378 Rust tests | 100% Core Coverage (322/535 files; optional/experimental excluded) | 109 Rust Neuron Models | 80-Model NetworkRunner
<p align="center">
<img src="docs/assets/spike_raster.png" width="800" alt="LIF spike raster — 5 neurons, sinusoidal input">
</p>
SC-NeuroCore is an open-source stochastic computing SNN framework
with FPGA synthesis. 116 neuron models (107 biophysical + 9 AI-optimized) spanning
82 years of computational neuroscience (McCulloch-Pitts 1943 through
ArcaneNeuron 2026) run inside a deterministic stochastic computing engine
with bit-true Verilog RTL co-simulation, FPGA synthesis via an IR compiler
(SystemVerilog + MLIR/CIRCT backends), an equation-to-Verilog compiler
that turns arbitrary ODE strings into synthesizable Q8.8 fixed-point RTL,
formal verification (7 SymbiYosys
modules, 67 properties), a Rust SIMD engine **39–202× faster than Brian2**
(27.7 billion synaptic events/s at 100K neurons, 109 Rust neuron models
with PyO3 bindings, 80-model NetworkRunner with Rayon-parallel populations),
CuPy GPU acceleration, JAX JIT training,
MPI distributed simulation (billion-neuron scale via mpi4py),
an identity continuity substrate (persistent spiking networks with
checkpointing and L16 Director control), a 128-function spike train
analysis toolkit (23 modules), 12 visualization plots, 13 advanced
plasticity rules (pair/triplet/voltage STDP, BCM, BPTT, TBPTT, EWC,
e-prop, R-STDP, MAML, STP, structural plasticity), 7 biological circuit
primitives (gap junctions, tripartite synapse, Rall dendrite, cortical
column, lateral inhibition, WTA, gamma oscillation), 10 model zoo
configurations with 3 pre-trained weight sets, 9 hardware chip emulators,
quantum hybrid computing (Qiskit + PennyLane + SC-to-quantum compiler),
surrogate gradient training reaching 99.49% MNIST accuracy, a
[NIR](https://neuroir.org/) bridge — FPGA backend for the
neuromorphic intermediate representation standard (18/18 primitives,
recurrent edges, multi-port subgraphs; verified interop with SpikingJelly,
snnTorch, and Norse), a SpikeInterface adapter for experimental data import,
ANN-to-SNN conversion (trained PyTorch models to rate-coded SNNs in one call),
trainable per-synapse delays (DelayLinear with differentiable interpolation),
one-command FPGA synthesis (`sc-neurocore deploy model.nir --target ice40` auto-runs Yosys+nextpnr+icepack if installed; generates project files for Vivado targets),
per-layer adaptive bitstream length for mixed-precision SC networks,
event-driven FPGA RTL (AER encoder, event neuron, spike router —
15-39x fewer register toggles than clock-driven at 0.01-10% activity, measured),
and a 6-codec neural data compression library (ISI, predictive, delta, streaming,
AER) with a unified API and auto-recommendation engine — targeting BCI
implants (Neuralink-scale 1024+ channels), neural probes (Neuropixels),
neuromorphic inter-chip routing, and real-time closed-loop telemetry.
2 928 passing Python tests and 378 Rust tests hold 100% core coverage.
13 CI workflows guard every push. conda-forge recipe ready.
## Feature Comparison
| Feature | SC-NeuroCore | snnTorch | Norse | Lava | Brian2 |
|---------|:---:|:---:|:---:|:---:|:---:|
| Stochastic computing (bitstream) | **Yes** | — | — | — | — |
| Bit-true RTL co-simulation | **Yes** | — | — | — | — |
| Verilog / FPGA synthesis | **Yes** | — | — | Loihi only | — |
| IR compiler → SystemVerilog | **Yes** | — | — | — | — |
| Rust engine (39–202× vs Brian2) | **Yes** | — | — | — | — |
| Surrogate gradient training | **7 surrogates, 10 cells** | Yes | Yes | Yes | — |
| PyTorch `nn.Module` SNN | **Yes** (+ SC weight export) | Yes | Yes | — | — |
| GPU acceleration | PyTorch + CuPy | PyTorch | PyTorch | — | — |
| Neuron model library | **116** | 11 | 6 | 3 | ~5 builtin |
| Rust neuron models (PyO3) | **109** | — | — | — | — |
| NetworkRunner (fused loop) | **80 models** | — | — | — | — |
| Network simulation engine | **3 backends** | PyTorch | PyTorch | Lava | C++ codegen |
| MPI distributed simulation | **Yes** | — | — | — | — |
| Pre-trained model zoo | **10 configs, 3 weights** | — | — | — | — |
| Spike train analysis | **128 functions** | — | — | — | — |
| Visualization plots | **12** | — | — | — | — |
| Advanced plasticity rules | **13** | — | — | — | — |
| Biological circuits | **7** | — | — | — | — |
| SC→quantum compiler | **Yes** | — | — | — | — |
| Predictive coding (SC) | **Yes** | — | — | — | — |
| Fault tolerance benchmark | **Yes** | — | — | — | — |
| Phi* (IIT) estimation | **Yes** | — | — | — | — |
| SpikeInterface adapter | **Yes** | — | — | — | — |
| NIR primitives | **18/18** | — | 12 | 5 | — |
| MNIST accuracy (SNN) | **99.49%** | ~95% | ~93% | — | — |
| Plasticity (STDP, R-STDP) | Yes | — | Yes | Yes | Yes |
| Quantum hybrid (Qiskit/PennyLane) | **Yes** | — | — | — | — |
| MLIR emitter (CIRCT) | **Yes** | — | — | — | — |
| Hyperdimensional computing | Yes | — | — | — | — |
| Formal verification (SymbiYosys) | **7 modules, 67 props** | — | — | — | — |
| JAX JIT training | **Yes** | — | — | — | — |
| CuPy sparse GPU | **Yes** | — | — | — | — |
| AI-optimized neurons | **9 (ArcaneNeuron + 8)** | — | — | — | — |
| Identity substrate | **Yes** | — | — | — | — |
| ANN-to-SNN conversion | **Yes** | — | — | — | — |
| Trainable per-synapse delays | **Yes** | — | — | — | — |
| One-command FPGA deploy CLI | **Yes** | — | — | — | — |
| Per-layer adaptive bitstream | **Yes** | — | — | — | — |
| Event-driven FPGA RTL (AER) | **Yes** | — | — | — | — |
| Raw waveform compression (24x) | **Yes** | — | — | — | — |
| Spike codec library (6 codecs) | **Yes** | — | — | — | — |
| Visual SNN Design Studio | **Yes (web IDE)** | Basic GUI | Jupyter | — | — |
| conda-forge recipe | **Ready** | Yes | — | — | Yes |
| PyPI package | Yes | Yes | Yes | Yes | Yes |
| License | AGPL-3.0 | MIT | LGPL-3.0 | BSD-3 | CeCILL-2.1 |
- **125-function spike train analysis toolkit** — CV, Fano factor, cross-correlation, Victor-Purpura distance, SPIKE-sync, Granger causality, GPFA, SPADE pattern detection, and 115 more functions. Matches Elephant + PySpike combined. Pure NumPy.
- **Neural data compression library** — Two layers: **WaveformCodec** compresses raw 10-bit electrode waveforms end-to-end (spike detection + template matching + LFP compression, 24x on 1024-channel Neuralink-scale data, fits Bluetooth uplink). **Spike raster codecs** (ISI+Huffman, Predictive with 4 learnable predictors, Delta, Streaming, AER) compress binary spike trains 50-750x. Unified API: `get_codec(name)`, `recommend_codec()`. Learnable world-model predictor (99.6% accuracy). Rust backend (780x speedup). Bit-true LFSR matches Verilog RTL.
SC-NeuroCore's niche: **deterministic stochastic computing with FPGA co-design** — Python simulation matches synthesisable RTL bit-for-bit (deterministic LFSR seeds, Q8.8 fixed-point, cycle-exact co-simulation).
### Performance: Rust Engine vs Brian2 (Brunel AI network)
Measured on i5-11600K @ 3.90 GHz, 300 ms simulation, 10% connection probability.
Stored artifact: [`benchmarks/results/rust_scaling_benchmark.json`](benchmarks/results/rust_scaling_benchmark.json)
| Scale | SC Rust | Brian2 | Speedup | SC synaptic events/s |
|------:|--------:|-------:|--------:|---------------------:|
| 1K neurons | 0.029 s | 2.689 s | **93×** | 110 M/s |
| 5K neurons | 0.285 s | 4.681 s | **16×** | 288 M/s |
| 10K neurons | 0.172 s | 6.754 s | **39×** | 1.86 B/s |
| 50K neurons | 0.582 s | 31.03 s | **53×** | 13.9 B/s |
| 100K neurons | 1.153 s | 232.3 s | **202×** | 27.7 B/s |
SIMD primitives: 190 Gbit/s popcount (AVX-512 dispatch, Criterion benchmark:
[`benchmarks/results/criterion_bitstream_2026-03-26.json`](benchmarks/results/criterion_bitstream_2026-03-26.json))
### Network Simulation Engine
Population-Projection-Network architecture with 3 backends:
| Backend | Scope | Performance |
|---------|-------|-------------|
| **Python** | Any of 116 neuron models | NumPy vectorized |
| **Rust NetworkRunner** | 80 models in fused Rayon-parallel loop | 100K+ neurons, near-linear scaling |
| **MPI** | Billion-neuron distributed simulation via mpi4py | Multi-node HPC clusters |
6 topology generators (random, small-world, scale-free, ring, grid, all-to-all),
12 visualization plots (raster, voltage, ISI, cross-correlogram, PSD, firing rate,
phase portrait, population activity, instantaneous rate, spike train comparison,
network graph, weight matrix), and 13 advanced plasticity rules (pair/triplet/voltage
STDP, BCM, BPTT, TBPTT, EWC, e-prop, R-STDP, MAML, homeostatic, STP, structural).
### Model Zoo
10 pre-built network configurations (Brunel balanced, cortical column, CPG,
decision-making, working memory, visual cortex V1, auditory processing, MNIST
classifier, SHD speech classifier, DVS gesture classifier) with 3 pre-trained
weight sets (MNIST 784-128-10, SHD 700-256-20, DVS 256-256-11).
### 116 Neuron Models (1943--2026)
Every model has a uniform `step(current) -> spike` API, a `reset()`, and a
cited reference. One file per model in `src/sc_neurocore/neurons/models/`.
| Category | Count | Examples |
|----------|------:|---------|
| Integrate-and-fire variants | 18 | AdEx, GLIF5, ExpIF, QIF, SFA, MAT, COBA-LIF, Parametric LIF, Fractional LIF |
| Simple spiking (2D+) | 20 | FitzHugh-Nagumo, Morris-Lecar, Hindmarsh-Rose, Resonate-and-Fire, Chay |
| Biophysical (conductance-based) | 20 | Hodgkin-Huxley, Connor-Stevens, Traub-Miles, Mainen-Sejnowski, Pospischil |
| Stochastic / population / neural mass | 13 | Poisson, GLM, Jansen-Rit, Wong-Wang, Wilson-Cowan, Ermentrout-Kopell |
| Rate / plasticity / other | 12 | McCulloch-Pitts (1943), Sigmoid Rate, Astrocyte, Amari, GatedLIF (2022) |
| Hardware chip emulators | 9 | Loihi CUBA, Loihi 2, TrueNorth, BrainScaleS AdEx, SpiNNaker, Akida, DPI |
| Multi-compartment | 7 | Pinsky-Rinzel, Hay L5 Pyramidal, Rall Cable, Booth-Rinzel, Dendrify |
| Map-based (discrete-time) | 6 | Chialvo, Rulkov, Ibarz-Tanaka, Cazelles, Courbage-Nekorkin, Medvedev |
| Core (stochastic computing) | 5 | StochasticLIF, FixedPointLIF, HomeostaticLIF, Dendritic, SC-Izhikevich |
| Training cells (PyTorch) | 4 | LIF, ALIF, RecurrentLIF, EProp-ALIF |
| **AI-optimized (novel)** | **9** | **ArcaneNeuron, MultiTimescale, AttentionGated, PredictiveCoding, SelfReferential, CompositionalBinding, DifferentiableSurrogate, ContinuousAttractor, MetaPlastic** |
### ArcaneNeuron — Self-Referential Cognition
The flagship AI-optimized model. Five coupled subsystems in a single ODE:
fast compartment (tau=5ms), working memory (tau=200ms), deep context
(tau=10s), learned attention gate, and a forward self-model (predictor).
The deep compartment accumulates identity: it changes only on genuine
novelty (prediction errors), not routine input. Confidence modulates
threshold and meta-learning rate. No equivalent in any other toolkit.
### Identity Substrate
Persistent spiking network for identity continuity (`sc_neurocore.identity`).
| Module | Class | Purpose |
|--------|-------|---------|
| `substrate.py` | `IdentitySubstrate` | 3-population network (HH cortical + WB inhibitory + HR memory) with STDP and small-world connectivity |
| `encoder.py` | `TraceEncoder` | LSH-based text-to-spike-pattern encoding |
| `decoder.py` | `StateDecoder` | PCA + attractor extraction + priming context generation |
| `checkpoint.py` | `Checkpoint` | Lazarus protocol: save/restore/merge complete network state (.npz) |
| `director.py` | `DirectorController` | L16 cybernetic closure: monitor, diagnose, correct network dynamics |
## Quick Start
```bash
pip install sc-neurocore
```
```python
from sc_neurocore import StochasticLIFNeuron
neuron = StochasticLIFNeuron(v_threshold=1.0, tau_mem=20.0, noise_std=0.0)
spikes = sum(neuron.step(0.8) for _ in range(500))
print(f"{spikes} spikes in 500 steps")
```
```bash
# Optional extras
pip install sc-neurocore[full] # all research modules
pip install sc-neurocore[gpu] # CuPy GPU acceleration
pip install sc-neurocore[nir] # NIR interop (Norse, snnTorch, Lava)
```
`pip install sc-neurocore` publishes the Python suite under the public
`sc-neurocore` package name. The optional Rust engine remains part of the
repository / release-asset / source-build flow rather than a separate PyPI
runtime dependency. Source-only Frontier modules such as `analysis`, `viz`,
`audio`, `dashboard`, and `swarm` still require a source checkout.
### Development Setup
```bash
git clone https://github.com/anulum/sc-neurocore.git
cd sc-neurocore
pip install -e ".[dev]" # editable install with all dev tools
make preflight # verify setup (lint + tests)
```
If you are changing the Rust bridge locally, install `bridge/` in the same
environment or run source-tree commands with `PYTHONPATH=src:bridge`.
### Visual SNN Design Studio (Experimental)
> **Status:** Development preview. The Studio is functional but under active
> development. API and UI may change between releases.
A web-based IDE for designing, training, compiling, and deploying spiking
neural networks — from ODE equations to FPGA bitstream in a single browser tab.
```bash
pip install sc-neurocore[studio]
sc-neurocore studio # opens browser at http://127.0.0.1:8001
```
| Feature | What it does |
|---------|-------------|
| **118 Model Browser** | Browse all neuron models by category, simulate with parameter sliders |
| **18+ Analysis Views** | Trace, phase portrait, ISI, f-I curve, bifurcation, heatmap, STA, frequency response, characterisation dashboard |
| **Compiler Inspector** | Build SC IR from equations, verify, emit SystemVerilog |
| **Synthesis Dashboard** | One-click Yosys synthesis to ice40/ECP5/Gowin/Xilinx, multi-target comparison, resource bars |
| **Training Monitor** | Live loss/accuracy curves via SSE, 6 surrogate gradients, per-layer spike rates |
| **Network Canvas** | Drag-and-drop populations and projections (React Flow), NIR export/import |
| **Full Pipeline** | Network → simulate → compile → synthesise in one click |
| **Project Save/Load** | Persistent workspaces as JSON, server-side storage |
No other SNN framework provides a visual design-to-hardware pipeline.
snnTorch has Jupyter notebooks. Brian2 has a basic GUI. Neither goes from
visual network design to FPGA resource estimation.
| Feature | SC-NeuroCore Studio | Brian2 GUI | snnTorch | Nengo GUI |
|---------|:---:|:---:|:---:|:---:|
| Visual network design | **Yes** | Basic | No | Yes |
| ODE equation editor | **Yes** | No | No | No |
| Live training curves | **Yes** | No | TensorBoard | No |
| Verilog output viewer | **Yes** | No | No | No |
| FPGA synthesis | **Yes** | No | No | No |
| Co-simulation view | **Yes** | No | No | No |
Full documentation: [Studio Guide](https://anulum.github.io/sc-neurocore/studio/)
## Docker
The Docker image ships with the full Rust engine (39–202× faster than Brian2):
```bash
# Build
make docker-build
# or: docker build -f deploy/Dockerfile -t sc-neurocore:latest .
# Run interactive Python shell
make docker-run
# or: docker run --rm -it sc-neurocore:latest
# Smoke test via docker compose
docker compose -f deploy/docker-compose.yml up
```
Pre-built images are published to GHCR on every release:
```bash
docker pull ghcr.io/anulum/sc-neurocore:latest
docker run --rm -it ghcr.io/anulum/sc-neurocore:latest
```
## Architecture
### Module Tiers
`pip install sc-neurocore` ships **Core + Simulation + Domain bridges** only.
Research and Frontier modules are available from source (`pip install -e ".[dev]"`).
| Tier | Modules | Ships in wheel | Status |
|------|---------|:--------------:|--------|
| **Core** | neurons, synapses, layers, sources, utils, recorders, accel, compiler, hdl_gen, hardware, cli, exceptions | Yes | Production-ready. 100% coverage. |
| **Simulation** | hdc, solvers, transformers, learning, graphs, ensembles, export, pipeline, profiling, models, math, spatial, verification, security | Yes | Stable. Import explicitly. |
| **Domain bridges** | quantum (Qiskit/PennyLane), adapters/holonomic (JAX), scpn (Petri nets) | Yes | Requires `pip install sc-neurocore[quantum]` or `[jax]` |
| **Research** | robotics, physics, bio, optics, chaos, sleep, interfaces | No | Tested. Available from source. |
| **Frontier** | generative, world_model, analysis, audio, dashboard, viz, swarm | No | Experimental. Available from source. |
| **Speculative** | `research/` (eschaton, exotic, meta, post_silicon, transcendent) | No | Theoretical. See `research/README.md`. |
### Architecture Diagram
```mermaid
graph TD
subgraph "Python API (pip install sc-neurocore)"
A[BitstreamEncoder] --> B[SCDenseLayer / SCConv2DLayer]
B --> C[116 Neuron Models<br/>LIF · HH · AdEx · Izhikevich · ArcaneNeuron · ...]
C --> NET[Network Engine<br/>Population · Projection · 3 Backends]
C --> ID[Identity Substrate<br/>Persistent SNN · Checkpoint · Director]
C --> D[STDP / R-STDP Synapses]
D --> E[BitstreamSpikeRecorder]
end
subgraph "Acceleration"
B --> F{Backend?}
F -->|CPU| G[NumPy / Numba SIMD]
F -->|GPU| H[CuPy CUDA]
F -->|Rust| I[sc_neurocore_engine<br/>39–202× vs Brian2 · 109 neuron models<br/>80-model NetworkRunner]
F -->|MPI| MPI[mpi4py distributed<br/>billion-neuron scale]
end
subgraph "Hardware Target"
I --> J[IR Compiler]
J --> K[SystemVerilog Emitter]
J --> K2[MLIR/CIRCT Emitter]
K --> L[Verilog RTL<br/>AXI-Lite + LIF Core]
K2 --> L
L --> M[FPGA Bitstream<br/>Xilinx / Intel]
L --> V[Formal Verification<br/>SymbiYosys · 7 modules]
end
subgraph "Domain Bridges (optional)"
B --> N[SCPN Petri Nets]
B --> O[Quantum Hybrid<br/>Qiskit / PennyLane]
B --> P[HDC/VSA Symbolic Memory]
end
style A fill:#2d6a4f,color:#fff
style I fill:#b5651d,color:#fff
style L fill:#1a237e,color:#fff
style M fill:#4a148c,color:#fff
style O fill:#6a1b9a,color:#fff
style V fill:#004d40,color:#fff
```
### Core API (28 symbols)
```python
from sc_neurocore import (
# Neurons
StochasticLIFNeuron, FixedPointLIFNeuron, FixedPointLFSR,
FixedPointBitstreamEncoder, HomeostaticLIFNeuron,
StochasticDendriticNeuron, SCIzhikevichNeuron,
# Synapses
BitstreamSynapse, BitstreamDotProduct,
StochasticSTDPSynapse, RewardModulatedSTDPSynapse,
# Layers
SCDenseLayer, SCConv2DLayer, SCLearningLayer,
VectorizedSCLayer, SCRecurrentLayer, MemristiveDenseLayer,
SCFusionLayer, StochasticAttention,
# Utilities
BitstreamEncoder, BitstreamAverager, RNG,
generate_bernoulli_bitstream, generate_sobol_bitstream,
bitstream_to_probability,
# Sources & Recorders
BitstreamCurrentSource, BitstreamSpikeRecorder,
)
```
### Hardware (Verilog RTL)
```
hdl/
sc_bitstream_encoder.v -- LFSR-based stochastic encoder (SEED_INIT param)
sc_bitstream_synapse.v -- AND-gate SC multiplier
sc_mux_add.v -- 2-input MUX (scaled addition)
sc_cordiv.v -- CORDIV stochastic divider (Li et al. 2014)
sc_dotproduct_to_current.v -- Popcount -> fixed-point current
sc_lif_neuron.v -- Q8.8 leaky integrate-and-fire
sc_firing_rate_bank.v -- Spike rate estimator
sc_dense_layer_core.v -- Full dense layer pipeline (decorrelated seeds)
sc_dense_matrix_layer.v -- N×M weight matrix layer
sc_axil_cfg.v -- AXI-Lite register file
sc_axil_cfg_param.v -- Parameterized AXI-Lite register file
sc_axis_interface.v -- AXI-Stream bulk bitstream I/O
sc_dma_controller.v -- DMA for weight upload and output readback
sc_cdc_primitives.v -- Clock domain crossing (2-FF sync, Gray, async FIFO)
sc_dense_layer_top.v -- Dense layer top wrapper
sc_neurocore_top.v -- System top (DMA + AXI + layers)
sc_aer_encoder.v -- AER spike encoder (event-driven output)
sc_event_neuron.v -- Event-triggered LIF (power ∝ spike rate)
sc_aer_router.v -- AER event distribution to target neurons
tb_sc_*.v (7 testbenches) -- Self-checking simulation testbenches
formal/ (7 modules) -- SymbiYosys formal verification properties
```
### GPU Acceleration
```python
from sc_neurocore.accel import xp, HAS_CUPY, to_device, to_host
from sc_neurocore.accel.gpu_backend import gpu_vec_mac
# VectorizedSCLayer auto-detects GPU
layer = VectorizedSCLayer(n_inputs=32, n_neurons=64, length=1024)
output = layer.forward(input_values) # GPU if CuPy available, else CPU
```
## Hardware-Software Co-Simulation
The co-sim flow verifies bit-exact equivalence between the Python model and
Verilog RTL:
```bash
# 1. Generate stimuli + expected results (Python golden model)
python scripts/cosim_gen_and_check.py --generate
# 2. Run Verilog simulation (requires Icarus Verilog)
iverilog -o tb_lif hdl/sc_lif_neuron.v hdl/tb_sc_lif_neuron.v
vvp tb_lif
# 3. Compare results
python scripts/cosim_gen_and_check.py --check
```
### Reproducibility
Every GitHub Release includes:
- **wheel + sdist** — Python distribution artifacts (`dist/sc_neurocore-*`)
- **SBOM** — CycloneDX software bill of materials (`sbom.json`)
- **Changelog extract** — release notes from `CHANGELOG.md`
Co-simulation traces are generated deterministically from fixed LFSR seeds.
To reproduce a published benchmark:
```bash
git checkout v3.13.3
pip install -e ".[dev]"
python benchmarks/benchmark_suite.py --markdown > BENCHMARKS.md
```
For Verilog co-sim trace reproduction, see `scripts/cosim_gen_and_check.py`
and the seed constants in `hdl/sc_bitstream_encoder.v`.
### Key Technical Details
- **LFSR**: 16-bit maximal-length, polynomial x^16+x^14+x^13+x^11+1, period 65535
- **Seed strategy**: Input encoders `0xACE1 + i*7`, weight encoders `0xBEEF + i*13`
- **Fixed-point**: Q8.8 (DATA_WIDTH=16, FRACTION=8), signed two's complement
- **Overflow**: Explicit bit-width masking via `_mask()` function
## Examples
Runnable scripts in `examples/`:
| Script | Description |
|--------|-------------|
| `01_basic_sc_encoding.py` | Bernoulli & Sobol bitstream encoding/decoding |
| `02_sc_neuron_layer.py` | SCDenseLayer construction, spike trains, and firing-rate summary |
| `03_ir_compile_demo.py` | IR graph building, verification, SystemVerilog emission (v3 Rust engine) |
| `04_vectorized_layer.py` | VectorizedSCLayer throughput benchmarking |
| `05_scpn_stack.py` | Full 7-layer SCPN consciousness stack with inter-layer coupling |
| `06_hdl_generation.py` | Verilog top-level generation from a network description |
| `07_ensemble_consensus.py` | Multi-agent ensemble orchestration and voting |
| `08_hdc_symbolic_query.py` | Hyper-Dimensional Computing symbolic memory (v3 Rust engine) |
| `09_safety_critical_logic.py` | Fault-tolerant Boolean logic with stochastic redundancy (v3 Rust engine) |
| `10_benchmark_report.py` | Head-to-head v2/v3 benchmark suite (v3 Rust engine) |
| `11_sc_training_demo.py` | Surrogate-gradient training of an SC dense layer (v3 Rust engine) |
| `12_load_pretrained_model.py` | Load pretrained ConvSpikingNet and classify MNIST digits |
| `jax_training_demo.py` | JAX JIT surrogate-gradient SNN training on synthetic data |
| `mnist_fpga/demo.py` | MNIST classifier: train → quantise Q8.8 → SC simulate → Verilog export |
| `mnist_conv_train.py` | **ConvSpikingNet: 99.49% MNIST** (learnable beta/threshold, cosine LR) |
| `mnist_surrogate/train.py` | Surrogate gradient SNN training (FastSigmoid/SuperSpike/ATan, ~95% MNIST) |
| `nir_roundtrip_demo.py` | NIR roundtrip: CubaLIF + recurrent connections, build → import → run → export |
| `norse_nir_roundtrip.py` | Norse → NIR → SC-NeuroCore roundtrip with real Norse weights |
| `snntorch_nir_roundtrip.py` | snnTorch RSynaptic → NIR → SC-NeuroCore roundtrip (CubaLIF + recurrent) |
| `spikingjelly_nir_roundtrip.py` | SpikingJelly → NIR → SC-NeuroCore roundtrip |
| `ann_to_snn_demo.py` | Convert trained PyTorch ANN to rate-coded SNN |
| `delay_training_demo.py` | Train spiking network with learnable per-synapse delays |
```bash
PYTHONPATH=src:bridge python examples/01_basic_sc_encoding.py
```
Examples marked **(v3 Rust engine)** require an available `sc_neurocore_engine`
bridge install. For source-tree runs against local bridge code, use
`PYTHONPATH=src:bridge` or install `bridge/` in the same environment.
## CI/CD
13 GitHub Actions workflows (`.github/workflows/`), all SHA-pinned:
| Workflow | Purpose |
|----------|---------|
| **ci.yml** | Lint (ruff format + ruff check + bandit) + Test (Python 3.10-3.14, coverage = 100%) + Build |
| **v3-engine.yml** | Rust engine `cargo test` + `cargo clippy` |
| **v3-wheels.yml** | Cross-platform wheels (Linux, macOS, Windows × Python 3.10–3.14) |
| **docker.yml** | Build & push Docker image to GHCR on release tags |
| **docs.yml** | MkDocs → GitHub Pages |
| **publish.yml** | Publish `sc-neurocore` to PyPI and `engine/` to crates.io on release tags |
| **release.yml** | Python wheel + sdist + changelog extraction → GitHub Release |
| **benchmark.yml** | Performance regression tracking |
| **codeql.yml** | CodeQL security analysis (weekly + on push) |
| **scorecard.yml** | OpenSSF Scorecard |
| **pre-commit.yml** | Pre-commit hook validation |
| **yosys-synth.yml** | Yosys HDL synthesis verification |
| **stale.yml** | Auto-label and close stale issues |
## Benchmarks
Run the benchmark suite:
```bash
python benchmarks/benchmark_suite.py # quick mode
python benchmarks/benchmark_suite.py --full # thorough (10x)
python benchmarks/benchmark_suite.py --markdown # output BENCHMARKS.md
```
Sample results (CPU, quick mode):
| Operation | Throughput |
|-----------|-----------|
| LFSR step | 2.25 Mstep/s |
| Bitstream encoder | 1.88 Mstep/s |
| LIF neuron step | 1.15 Mstep/s |
| vec_and (1024 words) | 45.67 Gbit/s |
| gpu_vec_mac (64x32x16w) | 6.15 GOP/s |
## Documentation
**Live site**: [anulum.github.io/sc-neurocore](https://anulum.github.io/sc-neurocore/)
- [Getting Started](docs/guides/getting-started.md) — Installation & quickstart
- [**Tutorials**](https://anulum.github.io/sc-neurocore/tutorials/01_stochastic_computing_fundamentals/) — 51 hands-on guides (SC fundamentals → MNIST → FPGA → quantum → formal verification)
- [API Reference](docs/api/API_REFERENCE.md) — Python package API
- [Rust Engine API](https://anulum.github.io/sc-neurocore/rust-api/sc_neurocore_engine/) — Rust engine docs
- [Hardware Guide](docs/hardware/HARDWARE_GUIDE.md) — FPGA deployment workflow
- [Architecture](docs/architecture/architecture.md) — Package architecture
- [Benchmarks](docs/benchmarks/BENCHMARKS.md) — Performance measurements
- [CHANGELOG.md](CHANGELOG.md) — Version history
Build docs locally:
```bash
pip install mkdocs mkdocs-material mkdocstrings[python]
mkdocs serve
```
## Install Extras
```bash
pip install sc-neurocore # core engine only (neurons, layers, compiler, HDL gen)
pip install sc-neurocore[accel] # + Numba JIT acceleration
pip install sc-neurocore[gpu] # + CuPy CUDA acceleration
pip install sc-neurocore[jax] # + JAX backend for holonomic adapters
pip install sc-neurocore[quantum] # + Qiskit + PennyLane quantum bridges
pip install sc-neurocore[lava] # + Intel Lava interop (Loihi target)
pip install sc-neurocore[research] # + matplotlib, networkx, onnx, torch
pip install sc-neurocore[full] # + numba, matplotlib, networkx, onnx, qiskit, pennylane
```
For development (includes all modules + research/frontier code from source):
```bash
pip install -e ".[dev]" # editable install with pytest, mypy, ruff, hypothesis
```
Pinned dependency files for reproducible environments:
```bash
pip install -r requirements.txt # runtime only
pip install -r requirements-dev.txt # runtime + dev tools
```
## Rust Engine (109 Neuron Models, 378 Tests)
The `sc_neurocore_engine` crate provides 109 Rust neuron models callable
from Python via PyO3 bindings (including ArcaneNeuron), an 80-model
NetworkRunner with Rayon-parallel population simulation (100K+ neurons),
and SIMD-accelerated primitives with dispatch across five ISAs (AVX-512,
AVX2, NEON, SVE, RISC-V V).
378 Rust tests across 17 test binaries.
| Category | Scope |
|----------|-------|
| Primitives | Bernoulli + Sobol bitstream, pack/unpack, popcount, SIMD (5 ISAs) |
| Neurons | 109 models: LIF variants, HH-type, maps, hardware emulators, population, ArcaneNeuron |
| NetworkRunner | 80-model fused simulation loop with CSR projections and Rayon parallelism |
| Synapses | Static, STDP, Reward-STDP |
| Layers | Dense, Conv2D, Recurrent, Learning, Fusion, Memristive, Attention |
| Networks | Brunel, GNN, Spike recorder, Connectome, Fault injection |
| Compiler | IR builder/parser/verifier, SystemVerilog + MLIR emitters, IR bridge |
| Domain | HDC, Kuramoto, SSGF geometry |
| Training | 6 surrogate gradient functions + property tests |
## Community
- [GitHub Discussions](https://github.com/anulum/sc-neurocore/discussions) — questions, ideas, show & tell
- [Issue Tracker](https://github.com/anulum/sc-neurocore/issues) — bug reports and feature requests
- [Contributing Guide](CONTRIBUTING.md) — how to set up, test, and submit PRs
## Citation
If you use SC-NeuroCore in your research, please cite:
```bibtex
@software{sotek2026scneurocore,
author = {Šotek, Miroslav},
title = {SC-NeuroCore: A Deterministic Stochastic Computing Framework for Neuromorphic Hardware Design},
version = {3.13.3},
year = {2026},
doi = {10.5281/zenodo.18906614},
url = {https://github.com/anulum/sc-neurocore},
license = {AGPL-3.0-or-later}
}
```
See also [`CITATION.cff`](CITATION.cff) for the machine-readable citation metadata.
## AI Disclosure
This project uses LLMs for advanced control mechanisms and GitHub
handling. All output is reviewed, tested, and verified by the project
author.
## License
SC-NeuroCore is dual-licensed:
- **Open Source**: [GNU Affero General Public License v3.0](LICENSE) (AGPLv3)
- **Commercial**: Proprietary license available for integration into closed-source products
For commercial licensing enquiries, contact [protoscience@anulum.li](mailto:protoscience@anulum.li).
---
<p align="center">
<a href="https://www.anulum.li">
<img src="docs/assets/anulum_logo_company.jpg" width="180" alt="ANULUM">
</a>
<a href="https://www.anulum.li">
<img src="docs/assets/fortis_studio_logo.jpg" width="180" alt="Fortis Studio">
</a>
<br>
<em>Developed by <a href="https://www.anulum.li">ANULUM</a> / Fortis Studio</em>
</p>