Duration: 18 months
The HPDP-40 is a unique a radiation hardened space qualified and high performance co-processor. Its key features are low power consumption, extreme flexibility and re-programmability on-the-fly, very low mass and compact form. Successful testing on Artificial Intelligence have been carried out on the HPDP with a limited scope, since there is no generic AI framework available for the HPDP. Klepsydra software products can be deployed on most used processors and operating systems and allow increasing data processing rate 10x and reduce energy consumption by 50%. Klepsydra Streaming Distribution Optimiser (SDO), part of Klepsydra AI, can find the optimal distribution of computing resources to execute AI models in the target processor, for the lowest latency, lowest power consumption or highest data rate. The SDO relies in Genetic Algorithms (GA) to find the optimal solution. The proposed idea consists of adapting Klepsydra AI and SDO for execution of AI on the HPDP. This combined hardware/software solution will address the current lack of space quantifiable AI inference solutions. The planned duration of this activity is 9 months, and planned to start in on 01.01.2023 and finished on 30.09.2023. The proposed idea is technically challenging due to the unique architecture of the HPDP co-processor and its programming language NML. Thus, there is a risk on how AI applications will be able to fit in the proposed solution, and it is a priority topic to be addressed. The main benefit of this proposed solution is the considerable simplification on the deployment of AI models in HPDP for two reasons: Klepsydra offers compatibility with mayor AI frameworks and no HPDP coding will be required. Finally, the proposed solution could enable AI applications on ESA future missions with high availability requirements (e.g. landers, deep space robotics, etc.). The proposed idea could be a hardware/software stack fully European that will strengthen Europe capability to use AI in space.
