Thesis defence L. Xie: logic design

26 February 2018 15:00 - Location: Aula, TU Delft - By: webredactie

Memristive Device for Logic Design and Computing. Promotor: Prof.dr.ir. S. Hamdioui (EWI).

Memristor is a promising emerging technology due to its good scalability, near-zero standby power consumption, high integration density, and CMOS fabrication compatibility. Several potential applications based on memristor technology have been proposed, such as non-volatile memories, neuromorphic systems, and resistive computing. However, research on resistive computing is still in its infancy phase. Therefore, it faces challenges with respect to development of device technology, logic design, computer architecture, compiler and applications.

This thesis focuses on logic design (including primitive logic gates, interconnect, circuit, and synthesis flow) based on memristor technologies and novel non-Von Neumann architecture.

Primitive logic gate – We first explore the complete logic gate space for Snider logic. Subsequently, we develop a novel logic design style referred to as scouting logic; it performs logic operations by modifying standard memory read operations. Although different logic design styles have been developed, innovative design styles still need to be explored. In addition, we analyze robustness of logic gates while considering memristive device variability, parasitic resistors and capacitors of nanowires, sneak path currents, and different memristor models. Despite the contribution of this thesis, more research is required in designing logic circuits that are robust against memristor device variability.

Interconnect design – We explore and compare three approaches to implement general interconnect schemes; they are using only memristor crossbar, only CMOS (i.e., metal wires and transistors) and a hybrid combination of both. We also propose a bus communication infrastructure at system level. In addition, we further explore the possibility to use dedicated interconnect schemes to address specific algebraic problems, such as matrix transpose. It worth to note that more efforts are required to generalize and optimize the communication infrastructure automatically.

Circuit design and synthesis flow – We develop methodologies to design ASICs and FPGAs using memristor logic design styles. For ASICs, we explore the place-and-route methods for large-scale circuits and develop an automated design and evaluation flow. For FPGAs, we develop two different implementations based on memristor logic, and we automate their design and evaluation flow. We observe that both ASICs and FPGAs based on memristor logic suffer from the CMOS control parts. An intelligent CMOS controller is therefore essential for overall improvements.

Non-Von Neumann architecture – We explore a micro non-von Neumann architecture, Computation-In-Memory (CIM Architecture), for specific data-intensive applications. CIM integrates both storage and processing elements in the same physical location using memristor technology; hence, it significantly reduces the memory access time and energy consumption. The preliminary results show that CIM architecture obtains significant improvements (e.g., energy-delay product and computational efficiency) over conventional multi-core architectures for specific applications (e.g., parallel addition, DNA sequencing, matrix multiplication).

More information?

For access to theses by the PhD students you can have a look in TU Delft Repository, the digital storage of publications of TU Delft. Theses will be available within a few weeks after the actual thesis defence.