.png){kind=icon}
.png){kind=icon}
Return
Dynamics analysis and FPGA implementation of a heterogeneous Hopfield neural network without self-feedback
.png){kind=icon}
PRE
AI
G
C
Y
H
X
Y
DOI：10.1007/s11071-026-12920-y.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
In Hopfield neural networks (HNNs), activation functions endow HNNs with intrinsic nonlinear characteristics. Heterogeneous activation functions can represent different types of neurons, thereby enhancing the biological plausibility of HNNs. However, HNNs with heterogeneous activation functions remain insufficiently investigated. To this end, this paper proposes a three-neuron heterogeneous HNN employing three different activation functions, namely arctangent, sine, and arcsine functions. The boundedness of the proposed network is theoretically proved, and the stability of its equilibrium points is analyzed. Numerical results demonstrate that the heterogeneous HNN exhibits complex dynamical behaviors, including chaotic orbits, periodic orbits, and stable points. Moreover, under different parameter conditions, two types of global coexisting attractors are observed, namely those composed of chaotic and periodic orbits and those composed of periodic orbits and stable points. In addition, an analog circuit is designed, and an FPGA-based hardware platform device is developed. Based on these implementations, PSIM circuit simulations and FPGA-based hardware experiments validate the numerical results, confirming the feasibility of the proposed network for neuromorphic hardware implementation.
Keywords:
Activation function
Coexistence attractors
Hopfield neural network
Hardware experiment
Journal
IF:
6
Papers:
1.4W
Citations:
4.1W
Organization
Cited Papers