Presentation
A Neuromorphic Quadratic, Integrate, and Fire Silicon Neuron with Adaptive Gain
Engineering in Medicine and Biology Conference (EMBC)
(2018)
Abstract
An integrated circuit implementation of a silicon neuron was designed, manufactured, and tested. The circuit was designed using the Quadratic, Integrate, and Fire (QIF) neuron model in 0.5 µm silicon technology. The neuron implementation was optimized for low current consumption, drawing only 1.56 mA per QIF circuit and utilized hysteretic reset, non-inverting integrator, and voltage-squarer circuits. The final area of each circuit in silicon was 268 µm height × 400 µm width. This design is the first IC of its kind for this neuron model and is successfully able to output true spiking that follows the behaviors of bistability, monotonic, and excitability spiking. The normal QIF design also features an easy way to change the time constant (which nominally operates in the millisecond range) of the spiking via a single, external capacitor (the only off-chip component in this design); the adaptive gain variation of the QIF circuit adds a second parameter that adjusts the time constant via an external resistor. The design also allows for an adjustable reset threshold and operates on a ±5 V power supply.
Keywords
- Modeling and identification of neural control using robotics,
- Biologically inspired robotics and micro-biorobotics - Biologically inspired locomotion
Disciplines
Publication Date
July, 2018
Location
Honolulu, HI
Citation Information
David W Parent and Eric J Basham. "A Neuromorphic Quadratic, Integrate, and Fire Silicon Neuron with Adaptive Gain" Engineering in Medicine and Biology Conference (EMBC) (2018) Available at: http://works.bepress.com/david_parent/44/