Methods for increasing the dependability of High-performance, Many-core, System-on-Chips

Future space exploration and exploitation missions will require significantly increased autonomy of operation for mission planning, decision-making, and adaptive control techniques. Spacecrafts will integrate new processing and compression algorithms that are often augmented with machine learning and artificial intelligence capabilities. This functionality will have to be provided with high levels of robustness, reliability, and dependability for conducting missions successfully. High-reliability requirements for space-grade processors have led to trade-offs in terms of costs, energy efficiency, and performance to obtain robustness. However, while high-performance / low-robustness configurations are acceptable in the Earth's vicinity, where assets remain protected by the planet's magnetosphere, they cease to work in more demanding environments, like cis-lunar or deep space, where high-energy particles will affect modern components heavily, causing temporary or permanent damage and ultimately system failures. The above has led to a situation where state-of-the-art processing elements (processors, co-processors, memories, special purpose accelerators, and field-programmable-gate arrays (FPGAs), all possibly integrated into System-on-a-Chip (SoC) designs) are superior to their high reliability, space-qualified counterparts in terms of processing power or energy efficiency. For example, from modern, state-of-the-art (SOTA) devices, one can expect a 2-3 order-of-magnitude performance per Watts improvement over space-grade equipment. Likewise, one finds a gap of approximately nine technology nodes between devices, which translates into a factor 25 decrease in operations per Watts. In this paper, we demonstrate how to utilize part of this enormous performance advantage to increase the robustness and resilience of otherwise susceptible semiconductor devices while harnessing the remaining processing power to build affordable space systems capable of hosting the compute-intensive functionality that future space missions require. We are bridging this performance-reliability gap by researching the enabling building blocks for constructing reliable and secure, space-ready Systems-on-a-Chip from SOTA processing elements.