Abstract
This paper mainly discusses a novel conceptual framework: fractional Hopfield neural networks (FHNN). As is commonly known, fractional calculus has been incorporated into artificial neural networks, mainly because of its long-term memory and nonlocality. Some researchers have made interesting attempts at fractional neural networks and gained competitive advantages over integer-order neural networks. Therefore, it is naturally makes one ponder how to generalize the first-order Hopfield neural networks to the fractional-order ones, and how to implement FHNN by means of fractional calculus. We propose to introduce a novel mathematical method: fractional calculus to implement FHNN. First, we implement fractor in the form of an analog circuit. Second, we implement FHNN by utilizing fractor and the fractional steepest descent approach, construct its Lyapunov function, and further analyze its attractors. Third, we perform experiments to analyze the stability and convergence of FHNN, and further discuss its applications to the defense against chip cloning attacks for anticounterfeiting. The main contribution of our work is to propose FHNN in the form of an analog circuit by utilizing a fractor and the fractional steepest descent approach, construct its Lyapunov function, prove its Lyapunov stability, analyze its attractors, and apply FHNN to the defense against chip cloning attacks for anticounterfeiting. A significant advantage of FHNN is that its attractors essentially relate to the neuron's fractional order. FHNN possesses the fractional-order-stability and fractional-order-sensitivity characteristics.
Introduction
Standard Hopfield Neural Networks (HNNs) are typically modeled using first-order differential equations, resulting in "memoryless" dynamics where the system's state depends solely on its immediate preceding state. However, many biological and physical processes exhibit non-local behavior and long-term memory. This paper introduces Fractional Hopfield Neural Networks (FHNN), which leverage fractional calculus to incorporate these history-dependent dynamics. By generalizing the HNN framework, the authors create a system where attractors are sensitive to the fractional order of the neurons, providing unique advantages for associative memory and hardware security.
Methodology and Implementation
The core of the FHNN is the "fractor," an analog circuit element that exhibits fractional-order impedance. Unlike standard resistors or capacitors, a fractor's behavior is governed by fractional derivatives, which are non-local operators.
Fractor Design
The authors implement the fractor using a self-similar, recursive lattice structure of resistors and capacitors. This net-grid configuration allows for the physical realization of fractional-order dynamics in an analog circuit.
Network Architecture
The FHNN is constructed by integrating these fractors into a recurrent neural network topology. Each neuron's state is determined by a fractional-order differential equation, and the network is trained using a fractional version of the steepest descent algorithm. This approach allows the system to seek fractional extreme points in its energy landscape.
Theoretical Stability and Attractor Analysis
A significant theoretical contribution of this work is the proof of Lyapunov stability for FHNNs. Using the Faà di Bruno formula, the authors construct a novel Lyapunov function to demonstrate that the system is asymptotically stable for fractional orders $v \in (0, 1)$.
A key finding is that the network's attractors are "order-sensitive." In a standard HNN, equilibrium points are determined by the weight matrix. In an FHNN, these points also depend on the specific fractional order $v$. For instance, in a two-neuron system, swapping the fractional orders between neurons results in a $\pi/2$ rotation of the equilibrium points, demonstrating high sensitivity to the underlying physical parameters.
Results and Applications
Simulations show that FHNNs possess enhanced associative capacity. A three-neuron FHNN can memorize up to eight distinct patterns, significantly outperforming its integer-order counterpart.
Furthermore, the sensitivity of the attractors to the fractional order makes FHNNs ideal for hardware security. Due to manufacturing tolerances, no two physical fractors will have identical fractional orders. This creates a unique, non-clonable "fingerprint" for each chip. A cloned chip with even minor variations in fractional order (e.g., $v=0.498$ vs $v=0.502$) will produce different attractor outputs, effectively defending against chip cloning attacks for anticounterfeiting.
Conclusion
FHNNs represent a significant advancement in recurrent neural network design by bridging fractional calculus and analog circuitry. While simulating these systems is computationally intensive due to their non-local nature, their unique properties—such as order-sensitive attractors and inherent long-term memory—open new avenues for high-capacity associative memory and robust hardware security. Future research will explore scaling these networks and investigating fractional chaotic dynamics.
Summary updated