| Summary: | Recently, thanks to the miniaturization and high performance of commercial-off-the-shelf<br />(COTS) computer systems, small satellites get popular. However, due to the very expensive launching<br />cost, it is critical to reduce the physical size and weight of the satellite systems such as cube satellites<br />(CubeSats), making it infeasible to install high capacity batteries or solar panels. Thus, the low-power<br />design is one of the most critical issues in the design of such systems. In addition, as satellites<br />make a periodic revolution around the Earth in a vacuum, their operating temperature varies greatly.<br />For instance, in a low earth orbit (LEO) CubeSats, the temperatures vary from 30 to -30 degrees<br />Celsius, resulting in a big thermal cycle (TC) in the electronic parts that is known to be one of the<br />most critical reliability threats. Moreover, such LEO CubeSats are not fully protected by active<br />thermal control and thermal insulation due to the cost, volume, and weight problems. In this<br />paper, we propose to utilize temperature sensors to maximize the lifetime reliability of the LEO<br />satellite systems via multi-core mapping and dynamic voltage and frequency scaling (DVFS) under<br />power constraint. As conventional reliability enhancement techniques primarily focus on reducing<br />the temperature, it may cause enlarged TCs, making them even less reliable. On the contrary,<br />we try to maintain the TC optimal in terms of reliability with respect to the given power constraint.<br />Experimental evaluation shows that the proposed technique improves the expected lifetime of the<br />satellite embedded systems by up to 8.03 times in the simulation of Nvidia’s Jetson TK1.
|
|---|