Challenges and solutions for large-scale integration of emerging technologies

Abstract

The semiconductor revolution so far has been primarily driven by the ability to shrink devices and interconnects proportionally (Moore's law) while achieving incremental benefits. In sub-10nm nodes, device scaling reaches its fundamental limits, and the interconnect bottleneck is dominating power and performance. As the traditional way of CMOS scaling comes to an end, it is essential to find an alternative to continue this progress. However, an alternative technology for general-purpose computing remains elusive; currently pursued research directions face adoption challenges in all aspects from materials, devices to architecture, thermal management, integration, and manufacturing. Crosstalk Computing, a novel emerging computing technique, addresses some of the challenges and proposes a new paradigm for circuit design, scaling, and security. However, like other emerging technologies, Crosstalk Computing also faces challenges like designing large-scale circuits using existing CAD tools, scalability, evaluation and benchmarking of large-scale designs, experimentation through commercial foundry processes to compete/co-exist with CMOS for digital logic implementations. This dissertation addresses these issues by providing a methodology for circuit synthesis customizing the existing EDA tool flow, evaluating and benchmarking against state-of-the-art CMOS for large-scale circuits designed at 7nm from MCNC benchmark suits. This research also presents a study on Crosstalk technology's scalability aspects and shows how the circuits' properties evolve from 180nm to 7nm technology nodes. Some significant results are for primitive Crosstalk gate, designed in 180nm, 65nm, 32nm, and 7nm technology nodes, the average reduction in power is 42.5%, and an average improvement in performance is 34.5% comparing to CMOS for all mentioned nodes. For benchmarking large-scale circuits designed at 7nm, there are 48%, 57%, and 10% improvements against CMOS designs in terms of density, power, and performance, respectively. An experimental demonstration of a proof-of-concept prototype chip for Crosstalk Computing at TSMC 65nm technology is also presented in this dissertation, showing the Crosstalk gates can be realized using the existing manufacturing process. Additionally, the dissertation also provides a fine-grained thermal management approach for emerging technologies like transistor-level 3-D integration (Monolithic 3-D, Skybridge, SN3D), which holds the most promise beyond 2-D CMOS technology. However, such 3-D architectures within small form factors increase hotspots and demand careful consideration of thermal management at all integration levels. This research proposes a new direction for fine-grained thermal management approach for transistor-level 3-D integrated circuits through the insertion of architected heat extraction features that can be part of circuit design, and an integrated methodology for thermal evaluation of 3-D circuits combining different simulation outcomes at advanced nodes, which can be integrated to traditional CAD flow. The results show that the proposed heat extraction features effectively reduce the temperature from a heated location. Thus, the dissertation provides a new perspective to overcome the challenges faced by emerging technologies where the device, circuit, connectivity, heat management, and manufacturing are addressed in an integrated manner.