No abstract

No abstract

I must take this opportunity to thank Prof. Christof Teuscher at Portland State University for his guidance and mentoring which helped me transform my research aspirations into a Master's Thesis. He has been a constant rock of support, and kept me focused and motivated throughout the period of this research. I might not have completed this work without his inspiring research methodology and approach to problem solving. The student group at Teuscher Lab has also been very supportive, and I am ever grateful to Mohammed, Walt, Wesley and Dat in particular for responding promptly whenever I had any questions. I also thank Prof. Mark Faust at PSU for his guidance and support that motivated me to pursue this thesis. I thank Prof. Marek Perkowski for being on my thesis committee and bringing his enthusiasm and inspiring personality to it. Last but not the least, the other source of constant support was my family. Hence, this thesis work is dedicated to my father, Ch Ramesh, my mother, Ch Madhavi, my two lovely sisters and the rest of my family.

We present a novel deep neural network (DNN) training scheme and RRAM in-memory computing (IMC) hardware evaluation towards achieving high robustness to the RRAM device/array variations and adversarial input attacks. We present improved IMC inference accuracy results evaluated on state-of-the-art DNNs including ResNet-18, AlexNet, and VGG with binary, 2-bit, and 4-bit activation/weight precision for the CIFAR-10 dataset. These DNNs are evaluated with measured noise data obtained from three different RRAM-based IMC prototype chips. Across these various DNNs and IMC chip measurements, we show that our proposed hardware noise-aware DNN training consistently improves DNN inference accuracy for actual IMC hardware, up to 8% accuracy improvement for the CIFAR-10 dataset. We also analyze the impact of our proposed noise injection scheme on the adversarial robustness of ResNet-18 DNNs with 1-bit, 2-bit, and 4-bit activation/weight precision. Our results show up to 6% improvement in the robustness to black-box adversarial input attacks.

No abstract