IMEC's neuromorphic chip is not the first that is able to self-train by association (see BrainChip provides details of neural network architecture ). However, it is novel in that it has used the variable resistance of memory devices to function in a manner similar to biological synapses.
The use of memory in this way – IMEC is researching the use of ReRAM and MRAM – offers up to two orders more energy efficiency than use of logic, according to An Steegan, executive vice president of semiconductor technology and systems at IMEC.
IMEC's goal is to develop manufacturing process technology and machine learning building blocks to make artificial intelligence sufficiently energy efficient that it can be integrated into sensors.
IMEC's chip has learned to compose music and the rules for the composition are learned on the fly.