Qualitative-Modeling-Based Silicon Neurons and Their Networks

Area of Science:

• Computational Neuroscience
• Neuro-inspired Engineering
• Integrated Circuit Design

Background:

• Complex ionic conductance models accurately simulate neuronal activity but are computationally intensive.
• Qualitative neuron models simplify these dynamics using low-dimensional polynomial differential equations, preserving core mathematical structures.
• These simplified models are crucial for bottom-up approaches in theoretical and computational neuroscience.

Purpose of the Study:

• To propose and review a qualitative-modeling-based approach for designing energy-efficient silicon neuron circuits.
• To demonstrate the implementation of these models in both analog and digital silicon circuits.
• To explore the application of these silicon neurons in neuromorphic systems like auto-associative memory.

Main Methods:

• Developing silicon-native differential equations that replicate the mathematical structures of polynomial-based qualitative neuron models.
• Designing CMOS analog silicon neuron circuits capable of diverse neuronal activities.
• Implementing digital silicon neuron circuits for Class I and II neuronal activity simulation.
• Constructing an auto-associative memory using a network of these silicon neurons.

Main Results:

• An analog silicon neuron circuit achieved various neuronal activities with power consumption under 72 nW, including square-wave bursting.
• Another analog circuit realized Class I and II neuronal activities at approximately 3 nW.
• A digital silicon neuron circuit successfully replicated Class I and II neuronal activities.
• Performance of an auto-associative memory was shown to be significantly influenced by the neuron class implemented in the network.

Conclusions:

• Qualitative modeling provides an effective strategy for designing low-power, configurable silicon neuron circuits.
• These silicon neurons, both analog and digital, can emulate diverse neuronal behaviors efficiently.
• The developed silicon neuron technology holds promise for advanced neuromorphic computing applications, such as associative memory.