Neuromorphic chip

from Wikipedia, the free encyclopedia

A neuromorphic chip is a microchip that is built based on the example of natural nerve networks . Neuromorphic chips are research areas in bionics and neuroinformatics . Existing prototypes include artificial retinas or even more advanced visual systems, artificial cochleae , odor detectors , or clock generators for natural motion sequences in robots . In particular, NPUs also belong to the class of neuromorphic microchips.

advantages

While in the Von Neumann architecture , the hardware very general and the software is therefore very held specifically to draw neuromorphic chips made by a highly specialized to the task hardware whose prominent feature high crosslinking (Inter connectivity ) with various feedback is. The aim is also to map the self-organizing development of the brain in circuits.

Neuromorphic electronics, which are characterized by high energy efficiency and small footprint compared to conventional chips , can probably be implanted in the future as a replacement for peripheral neuronal systems that have failed due to illness (e.g. a silicon retina for the retina of retinitis pigmentosa patients). Other areas of application are sensor systems for robots or other autonomous systems.

Neuromorphic chips are particularly suitable for pattern recognition and prediction of complex data.

disadvantage

Neuromorphic chips are physically deterministic , but this does not apply to data processing. Data is not saved, but learned and possibly forgotten. Neuromorphic circuits are also unsuitable for reliably and consistently performing precise numerical calculations.

See also

Individual evidence

  1. Christian Honey: Chips now with a brain. In: Technology Review. Heise, March 28, 2017, accessed July 25, 2017 .
  2. Hartmut Rehmsen: Neuromorphic Systems as the Coming AI Platform? In: AI Trendletter. SIGS DATACOM, accessed July 25, 2017 .

literature

Web links