A Mixed-Signal Structured AdEx Neuron for Accelerated Neuromorphic Cores

Area of Science:

• Neuroscience
• Neuromorphic Engineering
• Computational Neuroscience

Background:

• Existing neuromorphic hardware often uses simplified neuron models like the leaky integrate-and-fire (LIF) model.
• Cortical pyramidal neurons exhibit complex firing patterns and dendritic nonlinearities crucial for computation.
• Advanced hardware is needed to emulate these complex neuronal dynamics for more realistic simulations.

Purpose of the Study:

• To develop and characterize a multicompartment neuron circuit based on the adaptive-exponential integrate-and-fire (AdEx) model for the BrainScaleS hardware.
• To enable the emulation of complex neuronal behaviors, including sodium and calcium spikes and N-methyl-D-aspartate plateau potentials.
• To investigate how enhanced neuronal modeling on-chip improves computational capabilities.

Main Methods:

• Implementation of a multicompartment neuron circuit using the AdEx model on 65-nm CMOS technology.
• Leveraging an existing modular LIF architecture with extensions for exponential spike generation and conductance-based reset.
• Characterization of the AdEx circuit's ability to reproduce diverse firing patterns and emulate dendritic spikes.

Main Results:

• The AdEx circuit successfully reproduces a wide range of firing patterns observed in cortical pyramidal neurons.
• The circuit emulates sodium and calcium spikes, as well as N-methyl-D-aspartate plateau potentials.
• Demonstration of enhanced computational capabilities in single on-chip neurons due to nonlinear active signal processing.

Conclusions:

• The developed AdEx neuron circuit provides a more biologically realistic and computationally powerful platform for neuromorphic systems.
• This advancement facilitates the emulation of complex dendritic integration and spike generation in hardware.
• The interplay between passive and active signal processing in the circuit significantly boosts the computational power of individual neuromorphic neurons.