HENNC: Hardware Engine for Artificial Neural Network-based Chaotic Oscillators

Mobin Vaziri,Shervin Vakili,M. Mehdi Rahimifar,J.M. Pierre Langlois

2024-07-27

Abstract:This letter introduces a framework for the automatic generation of hardware cores for Artificial Neural Network (ANN)-based chaotic oscillators. The framework trains the model to approximate a chaotic system, then performs design space exploration yielding potential hardware architectures for its implementation. The framework then generates the corresponding synthesizable High-Level Synthesis code and a validation testbench from a selected solution. The hardware design primarily targets FPGAs. The proposed framework offers a rapid hardware design process of candidate architectures superior to manually designed works in terms of hardware cost and throughput. The source code is available on GitHub.

Hardware Architecture,Neural and Evolutionary Computing

What problem does this paper attempt to address?

The problem that this paper attempts to solve is to generate hardware cores for artificial neural network (ANN) - based chaotic oscillators in order to improve the throughput and hardware efficiency of pseudo - random number generators (PRNGs) in real - time operations. Specifically, the paper mainly focuses on the following aspects:

1. **Improving the efficiency of chaotic system modeling**: Although traditional numerical methods such as Runge - Kutta (RK) can accurately solve complex chaotic systems, their hardware implementation is complex and resource - consuming. The paper proposes using ANN to approximate chaotic systems, thereby simplifying hardware design and improving efficiency.

2. **Optimizing hardware architecture design**: Through an automated framework (HENNC), explore potential hardware architectures and generate corresponding synthesizable high - level synthesis (HLS) code and verification test benches. This enables designers to find the optimal hardware solution in a short time, reducing the time and cost of manual design.

3. **Enhancing security**: Chaotic oscillators are widely regarded as an ideal choice for constructing high - security PRNGs due to their complex dynamic characteristics, ergodicity, and sensitivity to initial conditions. The paper aims to use efficient hardware implementation to enhance the security of applications such as image encryption algorithms and resist cyber - attacks.

### Main contributions
- Proposed an automated framework for generating ANN - based chaotic oscillator hardware cores suitable for FPGA.
- Introduced novel delay and hardware cost estimation functions to accelerate design space exploration.
- Verified through experiments that the proposed framework is superior to manually - designed work in terms of hardware cost and throughput.

### Formula summary
The key formulas involved in the paper include:

- **Runge - Kutta 4th - order (RK - 4) method**:
\[
k_1^i = f_i(X_1, X_2, \ldots, X_N)
\]
\[
k_2^i = f_i\left(X_1+\frac{dt}{2}\cdot k_1^1, \ldots, X_N+\frac{dt}{2}\cdot k_1^N\right)
\]
\[
k_3^i = f_i\left(X_1+\frac{dt}{2}\cdot k_2^1, \ldots, X_N+\frac{dt}{2}\cdot k_2^N\right)
\]
\[
k_4^i = f_i\left(X_1+dt\cdot k_3^1, \ldots, X_N+dt\cdot k_3^N\right)
\]
\[
X_{i + 1}=X_i+\frac{1}{6}\times(k_1^i+2\times k_2^i+2\times k_3^i+k_4^i)
\]

- **ANN calculation formula**:
\[
Y_i^{(l)}=\phi\left(\sum_{j = 1}^{n^{(l-1)}}W_{ij}^{(l)}X_j^{(l-1)}+b_i^{(l)}\right)
\]
where $\phi$ is the activation function, $n$ is the number of neurons in the layer, and $w$ and $b$ are the weights and biases of the model respectively.

These formulas show how to describe the evolution of chaotic systems and the ANN calculation process through mathematical methods, and then guide the optimization of hardware design.