Emerging Technologies for Beyond Conventional Compute Systems

Abdelrahman G. M. A. Qoutb

Supervised by Eby Friedman

426 Computer Studies Building

https://rochester.zoom.us/j/96209504543 

The application of beyond CMOS technologies in innovative materials, memory, logic, and architectures will likely exhibit novel compute schemes and systems. This functional diversification, supported by beyond CMOS technologies, is expected to unleash a wide spectrum of novel solutions that have not been previously possible, such as reconfigurable nonvolatile logic and on-chip integrated sensor networks.

The primary objective of this dissertation is to bridge the gap among novel emerging devices, unconventional architectures, and computing schemes to support or replace conventional CMOS technologies to achieve next generation applications. In this dissertation, representative circuit and architectural advances are proposed that exploit the unique characteristics of emerging, beyond CMOS devices. The unique characteristics of these proposed systems include thermal sensitivity, non-volatility, extreme low power, and reconfigurability.

An MTJ is treated in this dissertation as an illustrative example of an emerging technology that can support beyond CMOS systems. MTJs are commonly used within commercial systems as an embedded memory. Importantly, MTJs are compatible with CMOS fabrication processes. In this dissertation, MTJs are proposed as a solution for several different compute schemes, including self-aware computers, compute in-memory, reconfigurable logic, and distributed compute systems.

In addition to these emerging technologies, superconductive electronics is considered as a standalone replacement for conventional CMOS systems. In superconductive electronics, one important logic family is based on single flux quanta (SFQ) to encode and process data.

Although beyond CMOS devices exhibit a wide range of functions that can replace or support conventional CMOS systems, superconductive devices also exhibit reliability issues that should be identified and addressed early in the technology development process. Since these devices suffer from low yield, advanced testability methodologies that target the unusual characteristics of the technology are required. Two different approaches are described in this dissertation to enhance the testability of SFQ systems. In the first approach, circuit solutions to enhance the controllability and observability of the internal nodes within SFQ systems are presented. In the second approach, high level fault models are proposed to characterize SFQ systems and generate the required test vectors to locate and identify Josephson junction-based faults.