Use this URL to cite or link to this record in EThOS:
Title: Precision-energy-throughput scaling of error tolerant signal processing applications
Author: Anam, M. A.
ISNI:       0000 0004 5359 0235
Awarding Body: University College London (University of London)
Current Institution: University College London (University of London)
Date of Award: 2014
Availability of Full Text:
Access from EThOS:
Full text unavailable from EThOS. Please try the link below.
Access from Institution:
The increases in power dissipation and transient fault rates experienced in applications based on state-of-the-art integrated processor technologies now lead to bottlenecks in energy consumption, processing-throughput scaling, and system reliability. Therefore, application and system designers are now beginning to accept the notion of error tolerance across the system stack, i.e., new methods that detect and tolerate (or mitigate) faults in exchange for resource efficiency at the application, runtime, compiler, architecture and hardware layers are now under intensive investigation. Within this context, this thesis proposes and develops novel approaches to data packing and numerical entanglement for accelerated, error-tolerant, signal processing applications. In particular, different data packing strategies suitable for integer and floating point inputs are studied, and they are subsequently used to create the new concept of numerical entanglement, which is proposed for highly-reliable numerical processing of integer data streams. The results of this thesis demonstrate that up to seven-fold decrease of processing time or energy consumption can be obtained if the signal processing application can operate under increased margins in the mean-squared error or signal-to-noise ratio of linear or sesquilinear processing kernels. In addition, for methods that scale processing throughput or energy consumption at the expense of reliability, our proposal for highly-reliable numerical processing of integer data streams is shown to detect all possible system-induced errors in one out of M input streams (M=3) with minimal overhead. Therefore, the thesis proposals can be used in software or hardware systems for resource scaling of error-tolerant signal processing applications.
Supervisor: Not available Sponsor: Not available
Qualification Name: Thesis (Ph.D.) Qualification Level: Doctoral
EThOS ID:  DOI: Not available