Novel electronic and spintronic devices for low power logic computation

Sayeef Salahuddin, Purdue University

Abstract

Power dissipation in CMOS devices has become one of the major challenges for scaling in the recent years. In fact, it has been suggested that the scaling may stop long before we reach any fundamental barriers because of our inability to remove the heat generated in these devices. This has stimulated significant efforts to look for novel devices that are fundamentally different in operating principle and dissipate much less power for each switching event. This thesis is geared toward this goal. We have proposed and analyzed novel devices for all three aspects of logic computing namely (i) switching (ii) memory and (iii) interconnects. In particular, we have proposed a novel concept showing that strongly interacting systems may allow switching with dissipation a few orders of magnitude lower than the conventional devices. We have established this concept theoretically by using both spintronic (magnets) and electronic (ferroelectric) examples. We have also looked at nano magnet based non volatile memory often called the Spin RAM. Because of the non-volatility, there is no static power dissipation in such devices making them extremely energy efficient compared to existing RAM technology. For this we have self consistently solved Non Equilibrium Green's Function (NEGF) for quantum transport with Landau Lifshitz Gilbert (LLG) Equation for magnetization dynamics. We have successfully explained the key experimental observations of MgO based Spin RAM devices namely switching current and associated TMR with the same set of parameters, a feat that has eluded simulation studies so far. We have also analyzed spin flip scattering in these devices and explained how it is consistent with experimental observations. Based on these experimental benchmarks, we have proposed novel device designs which should decrease the required switching current, which has been identified as the major challenge toward the commercialization of such devices. Finally, we have studied nanowires as possible interconnects for future. We have shown that these structures are fundamentally slow being crippled by the so-called 'kinetic inductance' that, we showed, is inversely proportional to the density of states of the material in consideration. Our predictions for carbon nanotubes in this regard were later confirmed by experiments.

Degree

Ph.D.

Advisors

Datta, Purdue University.

Subject Area

Electrical engineering

COinS