Neuromorphic technology is an essential enabler for cognitive computing. It is an architecture of memories which react to input patterns and can be compared to the brain because of its low power requirements, scalability, and instantaneous internal communications.
A neuromorphic chip is a fully parallel silicon neural network. It is a chain of identical elements (i.e., neurons) which can store and process information simultaneously. They are addressed in parallel and have their own “genetic” material to learn and recall patterns without running a single line of code and without reporting to any supervising unit.
Another result of the parallel architecture of the neuromorphic chip is its constant learning and recognition time regardless of the number of connected neurons, as well as the ability to expand the size of the neural network by cascading chips.
Neuromorphics are useful for domain-specific designs. Since semiconductors improvements are slowing (reflecting the end of Moore's Law), the peak power per mm of chip area is decreasing (due to the end of Dennard scaling), and the power budget per chip is not increasing (due to electro-migration and mechanical and thermal limits), and chip designers have already used multi-core (which is limited by Amdahl's Law), a path left for major improvements in performance-cost-energy is domain-specific architectures. They do only a few specialized tasks but do them extremely well.
Several companies have developed neuromorphic-enabled devices and systems, based on a variety of circuit designs, that support new applications including those that part of the emerging Internet of Things.
TechForecastDigest.com
-----------------------------------------------------------------------------------------------------------
Copyright © 2018 TechWise Media LLC All Rights Reserved
Komentáře