Applications

We develop energy-efficient and application-specific hardware solutions for future computing systems by utilizing the switching characteristics of emerging devices. 

Neuromorphic computing

Neuromorphic computing, inspired by the brain’s architecture, integrates memory and processing units, enabled by the ability of memristors to store data and perform computations within the memory unit. Neuromorphic computing research will be divided into two parts: volatile devices for neurons and non-volatile devices for synapses. Studying different orders of complexity is necessary for more complex neuronal behaviors, such as bursting. On the other hand, uniformity (low variability control) is a key requirement for synapses in order to exhibit highly linear and symmetric conductance updates. 

Hardware Security

Inherent variabilities can be used for hardware security solutions, such as physically unclonable functions (PUF) and true random number generator (TRNG). Unlike software-based security solutions, which rely on predictable algorithms and are vulnerable, hardware-based security solutions offer more robust protection against a wide range of threats due to their physical and stochastic characteristics. If security and computing functionalities coexist within a single hardware system, data encryption can be performed. In this way, we develop a tunable device that can perform homomorphic encryption, which is a form of encryption that allows one to perform computations directly on encryption data . 

Probabilistic computing

The big data era addressed by artificial intelligence requires a computing hardware capable of handling complex and large tasks, exceeding the conventional von Neumann architecture. While quantum computer shows the potential to exceed the conventional computer, its cryogenic operating temperature still remain an issue. On the other hand, probabilistic computing (p-computing) addresses the problems of the computing methods presented above. It uses probabilistic bits (p-bits), which produce ‘0’ and ‘1’ continuously changing over time, and p-bits have a probability of being ‘0’ and ‘1’ depending on an input variable that changes the probability. Volatile devices will extensively be studied for p-bit elements, and they will be used to form a p-computing network to solve optimization problems, such as integer factorization and traveling salesman problem. 

Digital memcomputing (Processing-in-memory)

Digital memcomputing is another type of in-memory computation using non-volatile devices capable of performing Boolean logic operations. It is categorized into stateful and nonstateful (or sequential) logic, depending on how logic input and output are represented. Stateful logic has an identical information carrier (resistance/conductance state in both input and output), while nonstateful logic adopts different information carriers (voltage input and resistance output). We propose device-specific logic schemes that enable efficient logic operations in terms of energy consumption, operation steps, and device count. 

Graph computing

Graph analytics in non-Euclidean domains are designed to interpret and analyze complex relationships, paths, and networks within data using nodes and edges. Memristor-based graph computing is an energy-efficient method to accelerate solutions of graph problems, which are computationally hard. Sneak paths and stochastic switching, two persistent challenges that impede the practical implementation of memristor crossbars, can be taken advantage for graph analysis. By mapping a graph network to a memristor crossbar array, different algorithms will be implemented for various applications, such as pathfinding, link prediction, and graphlet decomposition.

Active medium transmission: bio-inspired self-amplifying interconnects 

Interconnection lines generally require amplifiers during signal propagation due to signal loss caused by the metal line’s resistance. Lines are therefore divided into smaller dimensions, or new low-loss interconnect materials are required to meet the criteria for today’s electronic chips. Biological systems offer a framework that can address this issue through fundamentally different mechanisms. Leon Chua’s theory of local activity suggests that a component can amplify small perturbations at the Edge of Chaos (EoC). We explore various materials and structures to demonstrate axon-like self-amplifying interconnects.

Thermal computing 

Managing heat has been considered a major challenge in modern CMOS technology due to both static and dynamic power dissipations. On the other hand, there is a growing perspective that heat can serve as an information carrier (instead of being treated as a useless byproduct) in post-silicon devices, enabling new functionalities and on-chip energy recycling. With the concept of heat-assisted devices, we explore how such devices can facilitate more efficient computing.