Silicon Complementary-Metal-Oxide-Semiconductor (CMOS) technology underpins the hardware backbone for most computing functions to-date. However, in response to the growth of data-intensive analytic applications and the demand for greater processing power at reduced energy consumption, a dramatic shift in the computing paradigm beyond silicon CMOS is expected, which will be jointly fueled by material discovery, new device design, and innovative system architectures. In this regard, I will discuss two emerging device and material technologies.
The first part of the talk will address the remarkable recent progress in III-V channel MOSFET technology and its candidature as a drop-in replacement for silicon CMOS front-end microelectronics. I will describe the key enabling technology, the monolithic integration, new insights related to the quasi-ballistic electron transport, and the demonstration of record performing III-V metal-oxide-semiconductor-field-effect-transistors (MOSFETs) fabricated at MIT as part of my Ph.D. research.
The second part of the talk will depart from conventional CMOS approaches and introduce the use of oxide electronics for neuromorphic information processing. Using the insulator-metal transition (IMT) effect in in transition-metal oxides as an example, I will discuss the construction of an artificial integrate-and-fire neuron through a feedback-engineering of the intrinsic properties of IMT materials.
I will also describe the development of a complete electro-thermal device model for devices based upon the metal-to-insulator effect that accurately predicts the experimental performance of emerging devices such as artificial neurons and threshold selector switches (for emerging memory), and establishes the performance limits as well as design guidelines for such devices. Finally, by combining fundamental material and device understanding with new computing algorithms and non-classical computer architectures, I will outline some thoughts for future research directions in energy-efficient neuromorphic computing.