Skip to Main Content
Mobile Menu

A Memristor-Based Neuromorphic Computing Application

Masters Thesis, Electrical Engineering, Boise State University, December 2012

Adrian Rothenbuhler


Artificial neural networks have recently received renewed interest because of
the discovery of the memristor. The memristor is the fourth basic circuit element,
hypothesized to exist by Leon Chua in 1971 and physically realized in 2008. The
two-terminal device acts like a resistor with memory and is therefore of great interest
for use as a synapse in hardware ANNs. Recent advances in memristor technology
allowed these devices to migrate from the experimental stage to the application stage.
This Master’s thesis presents the development of a threshold logic gate (TLG),
which is a special case of an ANN, implemented with discrete circuit elements using
memristors as synapses. Further, a programming circuit is developed, allowing the
memristors and therefore the network to be reconfigured and trained in real-time. The
results show that memristors are indeed viable for use in ANNs, but are somewhat
hard to control as a lot of intrinsic device characteristics are still under investigation
and are currently not fully understood. A simple threshold logic gate was built and
can be reconfigured to implement AND, OR, NAND, and NOR functionality. The
findings presented here contribute towards improvements on the device as well as
algorithmic level to implement a memristor-based ANN capable of on-line learning.

Thesis in Scholarworks