As demand for artificial intelligence grows, researchers and technology companies are looking beyond earth for ways to power the massive computing infrastructure AI will require. Google’s Project Suncatcher envisions a network of interconnected, solar-powered satellites equipped with the company’s Tensor Processing Units (TPUs), a chip built specifically to perform AI computations.
As demand for artificial intelligence grows, researchers and technology companies are looking beyond earth for ways to power the massive computing infrastructure AI will require. Google’s Project Suncatcher envisions a network of interconnected, solar-powered satellites equipped with the company’s Tensor Processing Units (TPUs), a chip built specifically to perform AI computations.
Provisionally called “Fault-Aware AI for Spaceflight Applications,” the course will teach students how to train massive AI models on Google’s fastest supercomputers, Cloud TPUs, using the JAX coding framework. The course will be organized around six progressive modules exploring such areas as memory protection, silent data corruption, custom coding, and selective redundancy.
“Space computing and terrestrial AI infrastructure are a unified engineering challenge,” says Morrison. He explained that space is an extreme test case for reliable AI: if you can build an AI system that keeps working securely and safely despite radiation and hardware faults in space, then you’re learning techniques that can also work on earth for the benefit of all humanity. That makes this curriculum relevant to any engineer building reliable machine-learning systems at scale.”
In a capstone design project, students will test AI by conducting fault-injection experiments that simulate radiation-induced faults. Then, they will protect the AI through various “hardening,” or error resisting, strategies. Students will also write a professional report analyzing their test results and comparing their chosen design with published radiation data.
The curriculum Morrison will design through the grant is intended to have a broad reach. Individual modules can eventually be extracted as “plug-and-play labs” for integration into existing machine learning and systems courses. A veteran of the U.S. Navy, Morrison will develop a dedicated module with the Warrior-Scholar Project to extend these skills to veterans entering the space AI workforce. Additionally, a K–12 introductory module will be developed and disseminated through ChipsHub and GitHub, reaching high school students and educators globally.
An award-winning educator who has been active in expanding and training the next generation of computer science engineers, Morrison joined the Notre Dame faculty in 2019.