The NeuroMem CM1K chip can solve pattern recognition problems from text and data analytics, vision, audition, and multi-sensory fusion with orders of magnitude less energy and complexity than modern microprocessors.
The chip features 1024 neurons, all interconnected and working in parallel, recognizing one pattern among any number in a constant amount of time in the order of 500 nsec. The neurons behave collectively as a K-Nearest Neighbor classifier or a Radial Basis Function. They are trainable and especially suitable to cope with ill-defined and fuzzy data, high variability of context and even novelty detection. Last, but not least, multiple CM1K chips can be daisy-chained to scale a network from thousands to millions of neurons with the same simplicity of operation as a single chip.
Read more about CM1K, the home edition of a neuromorphic chip
The behavior and architecture of the NeuroMem neurons is highly inspired by the human brain
- Queries or stimuli are broadcasted to all the neurons at once which process them in parallel
- The winner takes all and inhibits weaker responders
- The neurons know when they do not know and can therefore learn
- The neurons auto-correct themselves if a teacher contradicts their original response
- The neurons operate at low frequency (Mhz) and are low power
- The number of neurons is highly scalable with no impact on the number of I/Os
- Finally, the neurons offer a feature which nature cannot reproduce yet and that is the ability to save and restore their knowledge.