ECIT's work on dependable systems nominated for 'Best Paper Award' at IEEE MICRO
A study that was conducted at ECIT for over a year was published and nominated for the 'Best Paper Award' at the prestigious IEEE/ACM International Symposium on Microarchitecture® (MICRO).
The IEEE/ACM International Symposium on Microarchitecture is the premier forum for presenting, discussing, and debating innovative microarchitecture ideas and techniques for advanced computing systems. The event is sponsored by the Institute for Electrical and Electronics Engineers (IEEE) and the Association for Computing Machinery (ACM) and several leading industries such as Intel, ARM, IBM and Microsoft.
Annually, highly ranked papers with innovative ideas are selected by the technical programme committee and are nominated for the best paper award. This year, ECIT's work was selected out of 446 reviewed papers and finally 82 accepted papers submitted from research groups across 25 countries.
The nominated article is entitled: 'DStress: Automatic Synthesis of DRAM Reliability Stress Viruses using Genetic Algorithms' and was co-authored by Dr. Lev Mukhanov and Dr Georgios Karakonstantis, at ECIT in collaboration with Prof Dimitrios Nikolopoulos, at Virginia Tech, USA.
The 'DStress' study presents a novel framework based on genetic algorithms that automatically searches and identifies data and memory access patterns that
make DRAM chips (used in most today’s processors) extremely prone to failures, without requiring any knowledge of the internal DRAM design, which is typically hidden. Identifying and even predicting the causes of such growing failures is extremely important for ensuring the secure, non-disruptive operation of portable and cloud, high-performance computing systems.
The application of the framework on 72 DRAM chips using a unique in- house experimental server at ECIT revealed that DStress discovers memory access and data patterns that induce many more failures than the traditionally used memory test micro-benchmarks. The developed framework opens up new avenues for identifying data and access patterns that make memories more prone to failures that is essential for developing new methods for identifying them early, avoiding them and for ultimately enhancing the dependability of future computing systems.
Dr Georgios Karakonstantis commented: ''We are extremely proud of our achievement in publishing a paper in this prestigious and highly competitive symposium this year.
In addition to publishing the paper, being nominated for the best paper award is an outstanding recognition for our work turning a rather ‘strange’ year due to the pandemic, into a special one.”