Tunable superconducting neurons for networks based on radial basis functions.

Area of Science:

• Superconducting electronics
• Artificial neural networks
• Quantum computing

Background:

• Superconducting technologies are crucial for high-performance, energy-efficient signal processing in niche applications.
• Neural networks, particularly those using radial basis functions, are powerful computational models.
• Implementing tunable activation functions is key for optimizing neural network performance.

Purpose of the Study:

• To investigate the fundamental components of superconducting neural networks utilizing radial basis functions.
• To analyze the static and dynamic activation functions of a proposed superconducting neuron model.
• To explore methods for tuning these activation functions to a Gaussian form with significant amplitude.

Main Methods:

• Theoretical examination of static and dynamic activation functions for superconducting neurons.
• Design and investigation of novel heterostructures for tunable inductors.
• Utilizing superconducting, ferromagnetic, and normal layers within heterostructures for adjustable inductance.

Main Results:

• Demonstrated the feasibility of implementing tunable activation functions in superconducting neural networks.
• Proposed and analyzed heterostructures capable of providing the necessary adjustable inductance.
• Characterized the static and dynamic behavior of the proposed neuron model with Gaussian activation.

Conclusions:

• Superconducting neural networks with tunable Gaussian activation functions are a promising direction for high-performance computing.
• The proposed heterostructure design offers a practical pathway for achieving the required tunability in superconducting circuits.
• Further research into these elements could unlock new possibilities in specialized hardware acceleration.