RESEARCH / FROM LEARNING TO HARDWARE
How can learning
become more efficient?
I explore this question across algorithms, emerging memory devices, and the systems that connect them.
01 / ALGORITHMS
Brain-inspired learning.
Understanding learning beyond conventional backpropagation is central to my research. I am interested in local learning rules and mechanisms that connect biological inspiration with practical neural networks.
Featured work: Forward Target Propagation
FTP forms layerwise targets using an additional forward pass and trains with local losses. The work explores how global error information can guide learning without symmetric backward weights.
TMLR · OCTOBER 2026
Publication and citation ↗02 / HARDWARE
Analog in-memory computing.
Emerging memory devices offer opportunities to bring computation closer to stored data. My interests connect device behavior, efficient computation, and learning algorithms that can work within hardware constraints.
Featured work: Distance-Aware In-Memory Softmax-Based Dense Associative Memory for energy-efficient analog retrieval
DIM-SDAM combines distance-aware associative retrieval with analog in-memory computing on 256 × 256 memristor crossbars. Unipolar inputs and single-step retrieval reduce hardware overhead and limit the accumulation of analog noise.
PREPRINT · UNDER REVIEW
Read the preprint ↗Featured work: Continuous Kinetic Model of Counterion-Mediated Redox Switching in 14-Bit Molecular Memristors for Analog In-Memory Computing
A continuous kinetic modeling framework connects counterion-mediated molecular switching with circuit simulation and analog in-memory computing. A Spectre-compatible compact model and vectorized Python solver enable device-to-system evaluation, including neural network integration.
UNDER REVIEW
Publication details ↗03 / SYSTEMS
Hardware–software co-design.
I approach algorithms and hardware as connected design choices. Efficient learning requires understanding both the computation a model needs and the physical system that carries it out.
DIM-SDAM is one example of this approach: adapting associative retrieval to the capabilities and constraints of memristor arrays. Explore the project ↑
Bridging disciplines
My background spans semiconductor materials, device fabrication, and brain-inspired learning. This perspective motivates work that connects physical constraints with computational goals.
Start a conversation ↗