Characterization and Testing of Neuromorphic Electronic Circuits
Faculty Mentor Information
Kurtis D. Cantley
Presentation Date
7-2017
Abstract
Neuro-inspired electronics are being designed to efficiently process complex information and accomplish tasks such as pattern recognition, object detection, and noise reduction and filtering. For this research, we perform electrical testing of complementary metal-oxide-semiconductor (CMOS) circuits that mimic the behavior of neurons and synapses in the brain. The neuron circuits are leaky integrate-and-fire type, generating voltage pulses of constant magnitude but with frequency that depends on the input stimulus. The leak rate, pulse width, and shape of the spike frequency versus input current curve are adjustable with control voltages. The synapse circuit, which controls the strength of a connection between two neurons, is a novel design that implements the spike timing-dependent plasticity (STDP) learning rule. It consists of several interconnected subcircuit blocks that we have tested individually. For both the neuron and synapse circuits, measured data is compared to simulation as well as biological measurements to demonstrate the effectiveness of the designs.
Characterization and Testing of Neuromorphic Electronic Circuits
Neuro-inspired electronics are being designed to efficiently process complex information and accomplish tasks such as pattern recognition, object detection, and noise reduction and filtering. For this research, we perform electrical testing of complementary metal-oxide-semiconductor (CMOS) circuits that mimic the behavior of neurons and synapses in the brain. The neuron circuits are leaky integrate-and-fire type, generating voltage pulses of constant magnitude but with frequency that depends on the input stimulus. The leak rate, pulse width, and shape of the spike frequency versus input current curve are adjustable with control voltages. The synapse circuit, which controls the strength of a connection between two neurons, is a novel design that implements the spike timing-dependent plasticity (STDP) learning rule. It consists of several interconnected subcircuit blocks that we have tested individually. For both the neuron and synapse circuits, measured data is compared to simulation as well as biological measurements to demonstrate the effectiveness of the designs.