Date

2015

Document Type

Dissertation

Degree

Doctor of Philosophy

Department

Electrical Engineering

First Adviser

Yan, Zhiyuan

Other advisers/committee members

Wagh, Meghanad D.; Li, Tiffany Jing; Annampedu, Viswanath

Abstract

As the technologies scaling down, more transistors can be fabricated into the same area, which enables the integration of many components into the same substrate, referred to as system-on-chip (SoC). The components on SoC are connected by on-chip global interconnects. It has been shown in the recent International Technology Roadmap of Semiconductors (ITRS) that when scaling down, gate delay decreases, but global interconnect delay increases due to crosstalk. The interconnect delay has become a bottleneck of the overall system performance. Many techniques have been proposed to address crosstalk, such as shielding, buffer insertion, and crosstalk avoidance codes (CACs). The CAC is a promising technique due to its good crosstalk reduction, less power consumption and lower area. In this dissertation, I will present analytical delay models for on-chip interconnects with improved accuracy. This enables us to have a more accurate control of delays for transition patterns and lead to a more efficient CAC, whose worst-case delay is 30-40% smaller than the best of previously proposed CACs. As the clock frequency approaches multi-gigahertz, the parasitic inductance of on-chip interconnects has become significant and its detrimental effects, including increased delay, voltage overshoots and undershoots, and increased crosstalk noise, cannot be ignored. We introduce new CACs to address both capacitive and inductive couplings simultaneously.Quantum computers are more powerful in solving some NP problems than the classical computers. However, quantum computers suffer greatly from unwanted interactions with environment. Quantum error correction codes (QECCs) are needed to protect quantum information against noise and decoherence. Given their good error-correcting performance, it is desirable to adapt existing iterative decoding algorithms of LDPC codes to obtain LDPC-based QECCs. Several QECCs based on nonbinary LDPC codes have been proposed with a much better error-correcting performance than existing quantum codes over a qubit channel. In this dissertation, I will present stabilizer codes based on nonbinary QC-LDPC codes for qubit channels. The results will confirm the observation that QECCs based on nonbinary LDPC codes appear to achieve better performance than QECCs based on binary LDPC codes.As the technologies scaling down further to nanoscale, CMOS devices suffer greatly from the quantum mechanical effects. Some emerging nano devices, such as resonant tunneling diodes (RTDs), quantum cellular automata (QCA), and single electron transistors (SETs), have no such issues and are promising candidates to replace the traditional CMOS devices. Threshold gate, which can implement complex Boolean functions within a single gate, can be easily realized with these devices. Several applications dealing with real-valued signals have already been realized using nanotechnology based threshold gates. Unfortunately, the applications using finite fields, such as error correcting coding and cryptography, have not been realized using nanotechnology. The main obstacle is that they require a great number of exclusive-ORs (XORs), which cannot be realized in a single threshold gate. Besides, the fan-in of a threshold gate in RTD nanotechnology needs to be bounded for both reliability and performance purpose. In this dissertation, I will present a majority-class threshold architecture of XORs with bounded fan-in, and compare it with a Boolean-class architecture. I will show an application of the proposed XORs for the finite field multiplications. The analysis results will show that the majority class outperforms the Boolean class architectures in terms of hardware complexity and latency. I will also introduce a sort-and-search algorithm, which can be used for implementations of any symmetric functions. Since XOR is a special symmetric function, it can be implemented via the sort-and-search algorithm. To leverage the power of multi-input threshold functions, I generalize the previously proposed sort-and-search algorithm from a fan-in of two to arbitrary fan-ins, and propose an architecture of multi-input XORs with bounded fan-ins.

Share

COinS