| Reliability of the components as well as interconnects has become a challenge in nanometer VLSI technology. In the 2007/2009 update, the International Technology Roadmap for Semiconductors (ITRS) explicitly calls achieving high reliability a big challenge for both CMOS and non-CMOS based designs for 22/16nm and beyond. Redundancy-based techniques have been widely used to correct faulty behavior of components and achieve high reliability. However, not much has been done for the interconnect part. N-tuple Modular Redundancy (NMR) systems, in particular, are all based on the majority voting. The voter unit, therefore, becomes a bottleneck for the correct operation of any NMR system. In this work, we first propose a novel current-based voting strategy to design a robust NMR system. We show that with this inexpensive strategy, we can completely eliminate the centralized voter unit and push NMR to the logic gate level. Our strategy achieves high reliability that is vital for future nanotechnology in which a high defect rate is expected. At the same time, it consumes less power and has less propagation delay compared to the conventional NMR systems. Second, we present a comprehensive reliability analysis of both the distributed NMR and conventional NMR systems. In doing so, we take the voter effect into account, historically omitted from such analysis. We show how our distributed voter resolves the scalability drawback of a centralized voter, and at the same time we offer a systematic way of doing tradeoff among reliability and area/power. This analysis is crucial for a designer for deciding how much overhead is needed to achieve a certain level reliability while keeping cost as low as possible. Third, we propose the concept of grid interconnect that establishes highly robust interconnects. We show that using a direct sequence spread spectrum communication strategy, namely Logic CDMA, and inexpensive transceivers, we can transfer data with extremely low error rates. Such a highly reliable communication network is vital for future nano-scale systems. Experimental results are reported to verify these three concepts, clarify the design procedures and measure the system's reliability. |
