Location
Room # EB204
Start Date
5-6-2019 9:10 AM
Description
Radiation from outer space can cause soft errors in microelectronic devices deployed at terrestrial altitudes on Earth. Cosmic rays entering the Earth’s atmosphere create a complex cascade of radioactive particles. The most likely form of cosmic radiation to cause soft errors in microelectronics at terrestrial levels are neutrons. SRAM-based FPGAs are susceptible to terrestrial cosmic ray induced soft errors. These soft errors occur infrequently for a single device deployed at terrestrial altitudes. When many FPGAs are deployed in a large-scale system, the impact of these soft errors on reliability can be significant. This study examines terrestrial cosmic ray induced soft errors and the effects they can have on large-scale deployment of FPGAs in cloud computing. Fifteen data-center-like designs were tested for sensitivity through fault injecting. Sensitivities ranged from less than 1% to about 12% of randomly injected faults resulting in unacceptable behavior. A hypothetical but realistic large-scale FPGA system, with 100,000 node deployed at a high-altitude, running the most sensitive design would experience the dominant failure mode of silent data corruption every 3.8 hours on average. This system would only be able to retain reliability level above 0.99 for about two minutes. Some soft error detection and recover approaches are discussed.
Terrestrial Cosmic Ray Induced Soft Errors and Large-Scale FPGA Systems in the Cloud
Room # EB204
Radiation from outer space can cause soft errors in microelectronic devices deployed at terrestrial altitudes on Earth. Cosmic rays entering the Earth’s atmosphere create a complex cascade of radioactive particles. The most likely form of cosmic radiation to cause soft errors in microelectronics at terrestrial levels are neutrons. SRAM-based FPGAs are susceptible to terrestrial cosmic ray induced soft errors. These soft errors occur infrequently for a single device deployed at terrestrial altitudes. When many FPGAs are deployed in a large-scale system, the impact of these soft errors on reliability can be significant. This study examines terrestrial cosmic ray induced soft errors and the effects they can have on large-scale deployment of FPGAs in cloud computing. Fifteen data-center-like designs were tested for sensitivity through fault injecting. Sensitivities ranged from less than 1% to about 12% of randomly injected faults resulting in unacceptable behavior. A hypothetical but realistic large-scale FPGA system, with 100,000 node deployed at a high-altitude, running the most sensitive design would experience the dominant failure mode of silent data corruption every 3.8 hours on average. This system would only be able to retain reliability level above 0.99 for about two minutes. Some soft error detection and recover approaches are discussed.
Comments
Session 3