Implementation of artificial neural networks into hardware: concepts and limitations

The implementation of artificial neural networks into hardware will only show their true potential since the networks need full parallel processing for real-time applications. By this way neural networks are a genuine challenge to microelectronics: not only many synapses have to be integrated with a high density but also many interconnections should be provided on many layers. First this paper describes the state-of-the-art and the potential of silicon technologies for artificial neural networks. Secondly the potential of nanoelectronics is demonstrated by some suggestions. For nanoelectronics one needs a specific system technique which gains the most of the advantages from the ultra large scale integration (ULSI). The key issue of this paper shows exemplarily how the Schrodinger equation from quantum mechanics can control both the characteristics of the devices and the algorithm of self-organization on system level, and the self-structuring of a system during implementation. Such differential equations seem to be the key algorithms for the ultimate hardware concepts in electronics.