NOTE! This site uses cookies and similar technologies.

If you not change browser settings, you agree to it. Learn more

I understand

Learn more about cookies at :

March 20, 2015 | New memory technologies for neuromorphic computing architectures

Artificial neural networks emulate biological synapses, storing “memories” of past events in the form of a physical property, conductance. The networks’ behavior makes it possible to establish a temporal correlation that is well suited to the processing of natural data like images and sounds.

Researchers successfully demonstrated the feasibility of integrating CBRAM components into CMOS circuits to create synapse matrices in 2012. In 2014 an experimental circuit was built, integrating two different synapse matrix structures (with and without an access transistor) and selection and control logic in 130 nm standard CMOS technology.

SynapsesIn initial testing of the experimental circuit, the artificial synapses behaved as predicted. However, the fabrication of embedded neuromorphic circuits will require the mastery of a broad array of new know-how including artificial synapse matrix design. Factors like the degree of precision of the synaptic values, the maximum matrix size, and the influence of parasite pathways will depend on two criteria: the technology used and the CMOS circuit’s on-resistance (RON).

The successful testing enabled CEA LIST to validate the researchers’ planned architecture strategies for the development of future embedded neuromorphic circuits.