#
π SiliconSignature
**Hardware-Bound Image Authentication for the AI Era**
*Prove image provenance with ASIC proof-of-work and Reed-Solomon watermarking*
[](LICENSE)
[](./Robust%20ASIC-Based%20Image%20Authentication%20Using%20Reed-Solomon%20LSB%20Watermarking.pdf)
[](https://silicon.p2pclaw.com)
[](https://silicon.p2pclaw.com)
[](https://github.com/Agnuxo1/siliconsignature-go)
[](https://github.com/Agnuxo1/siliconsignature-rust)
[](https://github.com/Agnuxo1/siliconsignature-ts)
[](https://github.com/Agnuxo1/silicon-android)
[](https://github.com/Agnuxo1/silicon-comfyui-node)
[](https://github.com/Agnuxo1/silicon-a1111-script)
[](https://github.com/Agnuxo1/silicon-browser-extension)
---
## π¨ The Problem
| Threat | Current Solutions | Why They Fail |
|--------|----------------|---------------|
| **Deepfakes** | Detection algorithms | Reactive β find fakes *after* creation |
| **Image forgery** | EXIF metadata | Stripped in 1 click |
| **AI watermarking** | C2PA, SynthID | Trusts corporations; removable |
| **NFT provenance** | Blockchain tokens | Just a URL, not the image |
**SiliconSignature is different.** We embed **unforgeable proof-of-work** directly into the image pixels β bound to a physical ASIC chip. No corporation to trust. No metadata to strip. No blockchain needed.
---
## β‘ How It Works
```
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β SIGN AN IMAGE β
β β
β 1. SHA-256 hash of image bytes β
β 2. Search for nonce via ASIC proof-of-work β
β 3. Reed-Solomon ECC encodes (hash + nonce + metadata) β
β 4. Embed in LSB of RGB channels (offset 0x20) β
β 5. Magic header "SSv1" + 5Γ redundancy β
β β
β Result: Image looks identical. But pixels carry proof. β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
β VERIFY AN IMAGE β
β β
β 1. Extract LSB from RGB channels β
β 2. Decode Reed-Solomon (tolerates 40% pixel loss) β
β 3. Validate nonce via SHA-256 check β
β 4. Confirm ASIC work was performed β
β β
β Result: Authentic or Tampered. Binary. No grey area. β
βββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
```
---
## π― Compared to Alternatives
| | **SiliconSignature** | **C2PA (Adobe)** | **SynthID (Google)** | **NFT** |
|---|:---:|:---:|:---:|:---:|
| **Open Source** | β
MIT | β Corporate | β Corporate | Varies |
| **No Dependencies** | β
Pure code | β Ecosystem | β API | β Blockchain |
| **Survives editing** | β
40% pixels | β οΈ Metadata only | β οΈ Compression | β None |
| **Hardware-bound** | β
ASIC PoW | β No | β No | β No |
| **Cost to forge** | **$10,000+** | $0 (strip metadata) | $0 (remove sig) | $0 (screenshot) |
| **Verification** | β
Offline | β Needs Adobe | β Needs Google | β Needs blockchain |
---
## ποΈ Ecosystem
This repository is the **central hub** β the original Python implementation + research paper. The ecosystem extends across 8 specialized repositories:
| Component | Language | Repository | Description |
|-----------|----------|------------|-------------|
| **PWA** | HTML/JS | [siliconsignature-web](https://github.com/Agnuxo1/siliconsignature-web) | Browser-based sign & verify |
| **CLI** | Go | [siliconsignature-go](https://github.com/Agnuxo1/siliconsignature-go) | Command-line tool + serverless |
| **Library** | Rust | [siliconsignature-rust](https://github.com/Agnuxo1/siliconsignature-rust) | Rust + WASM bindings |
| **Package** | TypeScript | [siliconsignature-ts](https://github.com/Agnuxo1/siliconsignature-ts) | npm package for Node/Browser |
| **Mobile** | Kotlin | [silicon-android](https://github.com/Agnuxo1/silicon-android) | Android camera app |
| **ComfyUI** | Python | [silicon-comfyui-node](https://github.com/Agnuxo1/silicon-comfyui-node) | Custom node for ComfyUI |
| **A1111** | Python | [silicon-a1111-script](https://github.com/Agnuxo1/silicon-a1111-script) | Script for AUTOMATIC1111 |
| **Browser** | JS | [silicon-browser-extension](https://github.com/Agnuxo1/silicon-browser-extension) | Chrome/Firefox extension |
---
## π Abstract
This project implements an image authentication system that combines ASIC-based proof-of-work (Antminer S9, BM1387) with LSB steganography protected by ReedβSolomon error-correcting codes. Signatures derived from deterministic PoW are embedded directly into pixel least-significant bits (LSBs) and protected using RS over GF(2^8). Experimental results show the watermark survives approximately 30β40% pixel destruction and still allows full signature recovery, comparable to QR Code Level H error correction.
**Keywords:** ASIC, Proof-of-Work, Image Authentication, ReedβSolomon, LSB Steganography, SHA-256, Antminer S9, Cryptographic Signatures, Error Correction
---
## π Introduction
Image authentication methods based on metadata (EXIF or PNG text chunks) are easily stripped by ordinary image processing tools. This work proposes a different approach that resists both metadata stripping and large-scale visual modification by combining:
- **Hardware-bound signatures:** PoW produced by the BM1387 ASIC (Antminer S9).
- **ReedβSolomon protected embedding:** RS codes over GF(256) applied to the signature payload.
- **Multi-layer redundancy:** Repetition of the encoded signature across the image to enable voting-based recovery.
### Contributions
- Complete system architecture for ASIC-based image authentication.
- Pure-Python ReedβSolomon encoder/decoder over GF(2^8).
- LSB steganography engine with configurable redundancy.
- Validation framework demonstrating 30β40% damage tolerance.
- Open-source implementation ready for deployment.
- Multi-platform ecosystem (PWA, CLI, Libraries, Mobile, Browser extensions).
---
## π§ Theoretical Background
### Proof-of-Work (PoW)
A PoW requires finding an input that produces a hash below a target: the system uses a double SHA-256 construction H(H(header || N)) < T (as in Bitcoin). The BM1387 ASIC deterministically produces valid nonces that prove significant hashing work.
### ReedβSolomon (RS)
ReedβSolomon codes RS(n,k) operate over GF(256) using the primitive polynomial:
`p(x) = x^8 + x^4 + x^3 + x^2 + 1 (0x11D)`
With n-k = 2t parity symbols, RS can correct up to t symbol errors.
### LSB Steganography
LSB embedding modifies the least significant bit of pixel color channels to store payload bits. Changes are per-channel Β±1 at most and are visually imperceptible.
---
## ποΈ System Architecture
The pipeline consists of four components:
1. **Image Hasher:** produces a SHA-256 hash of raw pixel bytes.
2. **ASIC Bridge:** submits the image hash as `prevhash` to a Stratum job; returns an ASIC-discovered valid nonce.
3. **RS Encoder:** encodes the signature payload using ReedβSolomon over GF(2^8).
4. **LSB Embedder:** writes the encoded, repeated signature into LSBs across the image channels.
**Diagram:** Image SHA-256 β ASIC Bridge (BM1387 PoW) β RS Encoder β LSB Embedder (Γrepeats) β Output
---
## π Signature Payload
| Field | Size | Description |
|:-------:|:--------:|:------------|
| hash | 64 bytes | SHA-256 hex digest of pixel data |
| nonce | 8 bytes | ASIC-discovered valid nonce |
| ntime | 8 bytes | Timestamp of mining operation |
| version | 8 bytes | Block version (e.g., 0x20000000) |
| status | ~24 bytes| Authentication status string |
---
## βοΈ Implementation Details
- **Pure Python ReedβSolomon:** GF(2^8) arithmetic using primitive polynomial 0x11D; log/antilog tables; generator polynomial generation for NS=32 parity symbols; RS(n, n-32) configuration.
- **Embedding parameters:** typical payload β 170 bytes (JSON), RS parity 32 symbols, and `SIGNATURE_REPEATS` = 5.
**Capacity analysis:** For an image WΓH with 3 channels, available LSB capacity = WΓHΓ3 bits. Example: 1280Γ720 β 2,764,800 bits; our ~8,200-bit payload uses ~0.3%.
---
## π Experimental Results
### Embedding Performance
| Image | Resolution | Payload | Capacity used | Time |
|:-----:|:----------:|:-------:|:-------------:|:----:|
| Imagen_test10.jpg | 494Γ493 | 8,200 bits | 1.12% | 0.3s |
| silicon_tv_v4.png | 1280Γ720 | 8,200 bits | 0.30% | 0.8s |
| test_4k.png | 3840Γ2160| 8,200 bits | 0.03% | 2.1s |
### Damage Tolerance (Vandalism Tests)
| Damage level | Copies recovered | Signature status |
|:------------:|:----------------:|:----------------:|
| 0% (Control) | 5/5 | VERIFIED |
| 10% (Minor) | 5/5 | VERIFIED |
| 20% (Moderate) | 4/5 | VERIFIED |
| 30% (Severe) | 3/5 | VERIFIED |
| 40% (Extreme) | 2/5 | VERIFIED |
| 50% (Critical) | 1/5 | MARGINAL |
A painted-shapes attack removing ~20% of pixels still allowed full signature recovery using RS and multi-copy voting.
---
## π οΈ Hardware Specifications
**Antminer S9 (BM1387)**
| Spec | Value |
|:----:|:-----:|
| Model | Antminer S9 |
| ASIC chip | BM1387 |
| Chip count | 189 |
| Hashrate (aggregate) | 14.0 TH/s |
| Power | 1,400 W |
| Efficiency | ~10B hashes/J |
**Comparison:** Intel i7-10700K β 21,000 hashes/J β ASIC β 533,000Γ more efficient for PoW.
---
## π Security Analysis
**Forgery resistance:** An attacker must either perform the PoW (requires ASIC hardware) or preserve enough LSB data despite RS limits. Reconstruction from partial data is prevented by RS syndromes.
| Attack | Mitigation | Effectiveness |
|:------:|:----------:|:-------------:|
| Metadata stripping | LSB embedding | Survives |
| Aggressive visual edits | RS + redundancy | Up to 40% recovery |
| Signature forgery | Hardware-bound PoW | Computationally infeasible |
| LSB destruction | 5Γ redundancy | Survives partial destruction |
**Limitations:** JPEG lossy compression and resizing destroy embedded LSB payloads; ASIC dependence (software-only simulation mode available in the [PWA](https://silicon.p2pclaw.com)).
---
## π Limitations & Future Work
- Support Extranonce2 embedding for full PoW verification without metadata.
- Implement a full BerlekampβMassey decoder to improve error correction.
- Add support for video frame authentication.
- Expand mobile verification via the [Android app](https://github.com/Agnuxo1/silicon-android).
- Integrate with popular image generators (ComfyUI, A1111, Fooocus, InvokeAI).
---
## π§ͺ Usage (Quick Start)
Run the repository scripts with Python 3:
- **Embed signature:** `python silicon_signature_engine.py `
- **Verify signature:** `python verify_silicon_art.py `
- **Simulate damage & recovery:** `python simulate_damage.py `
- **Utilities:** `manual_png_check.py`, `silicon_rs_watermark.py`, `asic_auth_portal.py`
For browser-based signing (no Python/ASIC required), use the **[Live PWA](https://silicon.p2pclaw.com)**.
---
## π Repository Contents
```
Secure_image_generation_with_ASIC_signature/
βββ asic_auth_portal.py # ASIC authentication portal
βββ silicon_signature_engine.py # Core signing engine
βββ verify_silicon_art.py # Verification tool
βββ silicon_rs_watermark.py # Reed-Solomon watermark module
βββ silicon_fog_analyzer.py # Fog/lighting analyzer
βββ silicon_fog_v4_session.py # Session management
βββ simulate_damage.py # Damage simulation tests
βββ manual_png_check.py # PNG validation utility
βββ Robust ASIC-Based Image # Full research paper (PDF)
β Authentication Using
β Reed-Solomon LSB Watermarking.pdf
βββ ASIC_Signature_Paper.html # HTML version of paper
βββ Secure_image_generation_with # Interactive HTML demo
β ASIC_signature.html
βββ benchmarks/ # Performance benchmarks
βββ dataset/ # Test image datasets
βββ docs/ # Documentation
βββ drivers/ # ASIC drivers
βββ Originals/ # Original test images
βββ proofs/ # Verification proofs
βββ tools/ # Additional tools
βββ .github/workflows/ # CI/CD workflows
βββ CITATION.cff # Citation metadata
βββ LICENSE # MIT License
βββ SECURITY.md # Security policy
βββ zenodo.json # Zenodo metadata
βββ README.md # This file
```
---
## π References
Selected references from the original manuscript:
- Katzenbeisser & Petitcolas, Information Hiding Techniques for Steganography and Digital Watermarking, 2000.
- Nakamoto, Bitcoin: A Peer-to-Peer Electronic Cash System, 2008.
- Reed & Solomon, Polynomial Codes Over Certain Finite Fields, 1960.
- NIST FIPS 180-4: Secure Hash Standard.
- Bitmain Antminer S9 documentation.
---
## π§βπ» Author & Contact
**Francisco Angulo de Lafuente** β Independent Researcher, Spain
- GitHub: https://github.com/Agnuxo1
- ResearchGate: https://www.researchgate.net/profile/Francisco-Angulo-Lafuente-3
- Kaggle: https://www.kaggle.com/franciscoangulo
- Hugging Face: https://huggingface.co/Agnuxo
- ORCID: 0009-0001-1634-7063
---
## π How to Cite
Please use the `CITATION.cff` file in this repository or cite as:
> Francisco Angulo de Lafuente (2026). *Secure Image Generation with ASIC Signature*. GitHub repository, v1.0.0. https://github.com/Agnuxo1/Secure_image_generation_with_ASIC_signature
---
## π License
[MIT License](LICENSE) Β© 2026 Francisco Angulo de Lafuente
---
**[π Live PWA](https://silicon.p2pclaw.com) Β· [π Paper](https://github.com/Agnuxo1/Secure_image_generation_with_ASIC_signature/blob/main/Robust%20ASIC-Based%20Image%20Authentication%20Using%20Reed-Solomon%20LSB%20Watermarking.pdf) Β· [π Issues](https://github.com/Agnuxo1/Secure_image_generation_with_ASIC_signature/issues)**