In a promising nanoelectronics device, namely, memristor based on metal oxides, there are many intermediate states with different conductivity between the limits of highly conductive and low-conducting states. These intermediate states can be used in the processes of associative learning of a neural network based on memristor synapses and simultaneous processing of input pulses, which consists in their weighing and summation in the neuroprocessor. By the method of simultaneous magnetron sputtering of two cathodes in a reactive oxygen environment, thin films of mixed oxides with a different mole ratio of titanium and aluminum were obtained. A method for obtaining a mixed oxide with a specified metal fractions by controlling the sputtering rates of cathodes using acoustic piezoelectric sensors is described. It is shown that the introduction of Al into titanium oxide improves the electrophysical characteristics of the memristor. The existence of an optimal fraction of Al dopant maximizing the memristor resistance ratio of the high-resistive and low-resistive states is established. The results indicate that the method of reactive magnetron deposition of mixed metal oxide by simultaneous sputtering of two cathodes provides a more uniform distribution of elements across the thickness of the active layer compared with the atomic layer deposition method. The uniform distribution is necessary to improve the stability of the memristor. It can be expected that in the memristors on mixed oxides TixSc1-xOy, HfxSc1-xOy, HfxY1-xOy, HfxLu1-xOy, ZrxSc1-xOy, ZrxY1-xOy, ZrxLu1-xOy an optimal dopant fraction corresponding to the maximally increased ratio of resistances in the high-resistance and low-resistance states will also be observed. Moreover, memristors on films with pure hafnium and zirconium oxides have a much larger range of resistive switching than titanium oxide.