Ms. Olga Krestinskaya

Graduate Student

King Abdullah University of Science and Technology (KAUST)

My research area is software-hardware co-design for in-memory computing architectures and Artificial Intelligence (AI) hardware. Mainly, I’m focusing on in-memory computing circuits and architectures for AI and neural network-based applications. My main Ph.D. project is Hardware-Aware Neural Architecture Search (NAS) algorithms for in-memory computing applications, which is a promising approach for software-hardware co-design and co-optimization of neural network models and hardware parameters.

During the internship at Duke University, I was focusing on in-memory computing architecture design for transformer-based neural networks, involving architecture-level optimization and mapping techniques for computer vision applications.

I’m also interested in memristor-based circuits and systems, and neuromorphic architectures. During my master’s studies, I was involved in the design and optimization of analog and mixed-signal peripheral circuits for different types of memristor-based neural networks, especially for low-power on-edge applications. I studied the effects of non-idealities of memristor devices, such as noise, variations, and design parameter mismatches, on the performance of these designs. I’ve also designed analog backpropagation learning circuits for analog-based neural networks.

In addition, I was also involved in projects related to quantized neural networks and brain-inspired algorithms.