Impact of Switching Variability, Memory Window, and Temperature on Vector Matrix Operations Using 65nm CMOS Integrated Hafnium Dioxide-based ReRAM Devices

Neuromorphic applications utilizing Non-Volatile Memory (NVM) devices have significant potential in imitating the functionality of biological synapses at lower power and area consumption than conventional CMOS. One such NVM candidate is Resistive Random Access Memory (ReRAM) devices utilizing transition metal oxides, such as hafnium oxide (HfO
2
) as the switching layer. These devices, while able to operate at multiple distinct levels, can suffer from key reliability issues. In this work we explored the use of our unique ReRAM-CMOS hybrid system to recognize images from the Caltech-101 dataset through image similarity processing. The approach utilized vector matrix multiplication (VMM) to correlate image similarity (or difference) between images. The effects of ReRAM switching variability, memory window (MW) and temperature were assessed for performance of VMM-based image recognition.