A Fault-Tolerant Interconnect Mechanism for NMR Nanoarchitectures

Electr. Eng. Dept., Univ. of Texas at Dallas, Richardson, TX, USA
IEEE Transactions on Very Large Scale Integration (VLSI) Systems (Impact Factor: 1.22). 11/2010; DOI: 10.1109/TVLSI.2009.2024779
Source: IEEE Xplore

ABSTRACT Redundancy techniques, such as N -tuple modular redundancy (NMR), has been widely used to correct faulty behavior of components and achieve high reliability. Almost all redundancy-based strategies rely on a majority voting. The voter, therefore, becomes a critical unit for the correct operation of any NMR system. In this paper, we propose a voterless fault-tolerant strategy to implement a robust NMR system design. We show that using a novel fault-tolerant communication mechanism, namely logic code division multiple access, we can transfer data with extremely low error rates among N modules and completely eliminate the need for a centralized voter unit. Such a highly reliable strategy is vital for future nanosystems in which high defect rate is expected. Experimental results are also reported to verify the concept, clarify the design procedure, and measure the system's reliability.

  • [Show abstract] [Hide abstract]
    ABSTRACT: High defect rates are associated with novel nanodevice-based systems owing to unconventional and self-assembly-based manufacturing processes. Furthermore, in emerging nanosystems, fault mechanisms and distributions may be very different from CMOS due to unique physical layer aspects, and emerging circuits and logic styles. Development of analytical fault models for nanosystems is necessary to explore the design of novel fault tolerance schemes that could be more effective than conventional schemes. In this paper, we first develop a detailed analytical fault model for the nanoscale application specific integrated circuits (NASIC) computing fabric and show that the probability of 0-to-1 faults is much higher than of 1-to-0 faults. We then show that in fabrics with unequal fault probabilities, using biased voting schemes, as opposed to conventional majority voting, could provide better yield. However, due to the high defect rates, voting will need to be combined with more fine-grained structural redundancy for acceptable yield. This entails degradation in performance (operating frequency) due to an increase in circuit fan-in and fan-out. We, therefore, introduce a new class of redundancy schemes called FastTrack that combine nonuniform structural redundancy with uniquely biased nanoscale voters to achieve greater yield without a commensurate loss in performance. A variety of such techniques are employed on a wire streaming processor (WISP-0) implemented on the NASIC fabric. We show that FastTrack schemes can provide 23% higher effective yield than conventional redundancy schemes even at 10% defect rates along with 79% lesser performance degradation.
    IEEE Transactions on Nanotechnology 07/2012; 11(4):720-730. · 1.80 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: As technology scales, the protection of Finite State Machines' (FSMs) states against single event upset (SEU) and multiple bit upsets (MBUs) becomes more difficult. In this paper, a self-checking approach to enhance the SEU/MBUs immunity of FSMs' states by replicating One-Hot code times for state encoding is presented. This approach can correct less than bit-flip faults in the state register per cycle. Bit-flips are treated as random events and modeled by applying Poisson distribution. Two characteristics of this approach are obtained through probability analysis: first, this approach performs better with the increase of , whereas worse when an FSM contains more states; second, this approach can offer more enhanced reliability than Binary or One-Hot state encoding with Triple Modular Redundancy (TMR). The former characteristic leads to the further improvement of this approach which is called state-reforming. The reliabilities of this proposed approach and its state-reformed solutions, as well as are all evaluated through simulations of fault injections.
    IEEE Transactions on Nuclear Science 10/2012; 59(5):2572-2579. · 1.22 Impact Factor
  • [Show abstract] [Hide abstract]
    ABSTRACT: Due to the small size of nanoscale devices, they are highly prone to process disturbances which results in manufacturing defects. Some of the defects are randomly distributed throughout the nanodevice layer. Other disturbances tend to be local and lead to cluster defects caused by factors such as layer misintegration and line width variations. In this paper, we propose a method for identifying cluster defects from random ones. The motivation is to repair the cluster defects using rectangular ranges in a range matching content-addressable memory (RM-CAM) and random defects using triple-modular redundancy (TMR). It is believed a combination of these two approaches is more effective for repairing defects at high error rate with less resource. With the proposed fault repairing technique, defect recovery results are examined for different fault distribution scenarios. Also the mapping circuit structure required for two conceptual 32×32 and 64×64 bit RAMs are presented and their speed, power and transistor count are reported.
    Electrical Engineering (ICEE), 2013 21st Iranian Conference on; 01/2013