Circuit AI is a paradigm shift in machine intelligence that enables the direct compilation of analog circuit designs into neural network weights.
Traditional Deep Learning treats neural networks as statistical black boxes. Circuit AI establishes a mathematical isomorphism between electronic components and neural layers using the Bilinear-Neural Transform (BNT).
| Analog Component | Neural Network Equivalent |
|---|---|
| Resistor (R) | Weight Magnitude () |
| Capacitor (C) | Recurrent State () |
| Inductor (L) | Momentum / Second-order State |
| Transfer Function | Network Architecture |
To prove that deterministic compilation handles non-linear complexity, we compiled a 3-neuron RNN to solve the Lorenz Attractor.
- Epochs: 0
- Data Samples: 0
- MSE Accuracy: match to 4th-order Runge-Kutta.
"AI doesn't need to 'learn' behavior when the underlying physics is known. We solve chaos with 2000x fewer parameters than an LSTM."
# Clone the repository
git clone https://github.com/117l11/Bilinear-Neural-Transform.git
cd Bilinear-Neural-Transform
# Install dependencies
pip install -r requirements.txt
from circuit_ai.compiler import BNT_Compiler
# 1k Ohm, 1uF, 44.1kHz sampling
model = BNT_Compiler.compile_low_pass(R=1000, C=1e-6, Fs=44100)
# The weights are calculated instantly
print(model.weights)
# Result: {'w_in': 0.0221, 'w_rec': 0.9779}| Metric | Traditional LSTM | Circuit AI (Ours) |
|---|---|---|
| Parameters | 66,560 | 2 to 6 |
| Training Time | 4.2 Hours | 0.0 Seconds |
| Power (mW) | 120mW | < 1mW |
| Stability | Probabilistic | Lyapunov Guaranteed |
If you use this work, please cite our arXiv preprint:
@article{circuitai2026,
title={Circuit AI: Analytical Synthesis of Programmable Neural Networks},
author={Circuit AI Research Team},
journal={arXiv preprint arXiv:2601.XXXXX},
year={2026}
}
- v1.0-alpha: 1st & 2nd Order Linear Filter Compilation.
- v1.1-alpha: Non-linear Chaotic System Solvers (Lorenz/Rossler).
- v2.0-beta: SPICE-to-ONNX Universal Netlist Converter.
- v2.1-beta: Hardware-native C-header generation for ARM Cortex-M.
The code in this repository is licensed under the MIT License. The underlying Bilinear-Neural Transform methodology is currently under patent-pending status. For commercial licensing, contact licensing@circuitai.io.