Abstract

Miniaturization of integrated circuits (ICs) due to the improvement in lithographic techniques in modem deep sub-micron (DSM) technologies allows several complex processing elements to coexist in one IC, which are called System-on-Chip. As a first contribution, this thesis quantitatively analyzes the severity of timing constraints associated with Clock Distribution Network (CDN) in modem DSM technologies and shows that different processing elements may work in different dock domains to alleviate these constraints. Such systems are known as Globally Asynchronous Locally Synchronous (GALS) systems. It is imperative that different processing elements of a GALS system need to communicate with each other through some interfacing technique, and these interfaces can be asynchronous or synchronous. Conventionally, the asynchronous interfaces are described at the Register Transfer Logic (RTL) or system level. Such designs are susceptible to certain design constraints that cannot be addressed at higher abstraction levels; crosstalk glitch is one such constraint. This thesis initially identifies, using an analytical model, the possibility of asynchronous interface malfunction due to crosstalk glitch propagation. Next, we characterize crosstalk glitch propagation under normal operating conditions for two different classes of asynchronous protocols, namely bundled data protocol based and delay insensitive asynchronous designs. Subsequently, we propose a logic abstraction level modeling technique, which provides a framework to the designer to verify the asynchronous protocols against crosstalk glitches. The utility of this modeling technique is demonstrated experimentally on a Xilinx Virtex-II Pro FPGA. Furthermore, a novel methodology is proposed to quench such crosstalk glitch propagation through gating the asynchronous interface from sending the signal during potential glitch vulnerable instances. This methodology is termed as crosstalk glitch gating. This technique is successfully applied to obtain crosstalk glitch quenching in the representative interfaces. This thesis also addresses the dock skew challenges faced by high-performance synchronous interfacing methodologies in modem DSM technologies. The proposed methodology allows communicating modules to run at a frequency that is independent of the dock skew. Leveraging a novel clock-scheduling algorithm, our technique permits a faster module to communicate safely with a slower module without slowing down. Safe data communications for mesochronous schemes and for the cases when communicating modules have dock frequency ratios of integer or coprime numbers are theoretically explained and experimentally demonstrated. A clock-scheduling technique to dynamically accommodate phase variations is also proposed. These methods are implemented to the Xilinx Virtex II Pro technology. Experiments prove that the proposed interfacing scheme allows modules to communicate data safely, for mesochronous schemes, at 350 MHz, which is the limit of the technology used, under a dock skew of more than twice the time period (i.e. a dock skew of 12 ns)