Submission deadline: CLOSED
Publication: July/August 2018
Guest Editors: Natalie Enright Jerger (University of Toronto) and Joshua San Miguel (University of Toronto/University of Wisconsin-Madison)
Approximate computing covers a broad spectrum of systems and architectures where quality (or accuracy) serves as a design parameter, enabling trade-offs between quality of results and efficiency. This computing paradigm has garnered much research activity in recent years for two reasons. First, the dark future of CMOS scaling has forced architects to come up with new ways to squeeze every last ounce of efficiency in their designs. Second, there has been growing interest in applications that are inherently probabilistic, imprecise, or noisy (such as machine learning, multimedia, and sensor devices).
Approximate computing introduces fundamentally new research avenues due to its unique design principles: 1) unlike when tuning for efficiency, the quality knob has strict constraints in order to maintain correctness in the system; and 2) users and programmers need to have more active roles in deciding when quality is acceptable. This special issue of IEEE Micro will explore exciting, new ideas in the vast design space of approximate computing.
Topics of interest include (but are not limited to):
- Architectural support for approximate computing
- Approximation techniques on emerging processor and memory technologies
- Design methodologies and tools for approximate hardware
- Analog and circuit-level approximation techniques
- Language, compiler, and operating system support for approximate architectures
- Hardware accelerators for approximation-tolerant application domains
- Techniques for monitoring and controlling approximation quality