Professor Shanshan Liu of New Mexico State University presented her research at the Vanderbilt Seminar Series.
Efficient yet reliable data processing in microelectronic systems is fundamental to meet the computational requirements found in safety-critical applications. Microelectronic systems are prone to soft errors in the hardware implementation, causing data corruption and then possibly leading to a system failure. This makes error tolerance a critical feature to enable a wide adoption of techniques (such as Machine Learning) in safety-critical applications. However, protection against soft errors typically incurs in a significant penalty in hardware resources, and it has become a significant challenge for today’s high-performance computing systems. This seminar focuses on efficient error tolerant approaches for data storage and data computation, namely feature-driven error correction codes and algorithm-based error tolerance, respectively. These approaches ensure reliable operation while substantially improving hardware metrics such as the redundancy and power dissipation for error protection.