Design and implementation of multipattern generators in analog VLSI

Area of Science:

• Computational biology
• Neuroscience
• Robotics

Background:

• Small neural networks with fixed connectivity can generate multiple output rhythms from transient inputs via simulation.
• These networks are hypothesized to be crucial for biological behaviors, including dynamic gait control.

Purpose of the Study:

• To present a novel method for designing continuous-time recurrent neural networks (CTRNNs) with multiple embedded limit cycles.
• To demonstrate the ability to switch between these limit cycles using simple transient inputs.
• To describe the design and testing of a four-neuron CTRNN chip for implementing neural network pattern generators.

Main Methods:

• Design of continuous-time recurrent neural networks (CTRNNs) with embedded limit cycles.
• Utilizing transient inputs to switch between embedded limit cycles.
• Implementation and testing of a four-neuron CTRNN chip for pattern generation.

Main Results:

• Successfully designed CTRNNs capable of generating multiple rhythmic outputs.
• Demonstrated effective switching between embedded limit cycles using transient inputs.
• Validated the performance of a four-neuron CTRNN chip, with measured waveforms closely matching numerical simulations.

Conclusions:

• The proposed method enables the design of CTRNNs with multiple controllable rhythmic patterns.
• The developed CTRNN chip effectively implements these multipattern generators.
• This work provides a foundation for applying CTRNNs in biological systems and robotics, such as dynamic gait control.