A reconfigurable on-line learning spiking neuromorphic processor comprising 256 neurons and 128K synapses

Area of Science:

• Neuroscience
• Computer Engineering
• Artificial Intelligence

Background:

• Compact, low-power artificial neural processing systems with real-time on-line learning are a significant challenge.
• Neuromorphic engineering seeks to emulate biological neural systems for advanced computation.

Purpose of the Study:

• To present a full-custom mixed-signal VLSI device with neuromorphic learning circuits.
• To enable exploration of computational neuroscience models and construction of brain-inspired computing systems.

Main Methods:

• Designed a mixed-signal VLSI device with 128K analog synapses and 256 neuron circuits.
• Emulated biophysics of spiking neurons and dynamic synapses with short-term and long-term plasticity.
• Integrated asynchronous digital logic for network configuration and property setting.

Main Results:

• The device supports on-chip configuration of recurrent and deep networks with plasticity.
• Fabricated prototype (180 nm CMOS) occupies 51.4 mm² and consumes ~4 mW.
• Demonstrated biologically plausible dynamics and bi-stable spike-based plasticity for on-line learning.

Conclusions:

• The developed neuromorphic chip facilitates the realization of intelligent autonomous systems.
• The device supports cortical-like computational modules and on-line learning capabilities.
• This work advances the field of brain-inspired computing systems.