Programmable bistable switches and resistors for neural networks
E. G. Spencer · AIP conference proceedings · 1986
A system of crystalline materials based on bismuth oxide is being evaluated for use as programmable bistable switches and resistors in neural networks for computing. Included are Bi2O3, Bi12GeO20, Bi12SiO2, and Bi4Ge3O12. Both switching and electrical programming of resistors have been observed with voltages as low as 50 mV and with currents in the nanoampere range. Generally the resistances are of the order of 106 to 109Ω in the OFF position, switching to the order of 104 to 105Ω in the ON position. Present efforts are directed toward improving the metallurgy of the cross stripe conductors and the quality of the films in order to be able to fabricate large scale arrays in which all of the elements have the same operating characteristics with resistances in the R=108 Ω to 1011 Ω range. Also, computer controlled programming and measurement methods are being implemented which will duplicate the exact operating conditions in neural network analog computation.