This $200,000 National Science Foundation project grant supports research at Purdue University to develop compiler-based and hardware solutions enabling caches and graphics processing units in energy harvesting systems. The goal is to design next-generation energy harvesting devices with high performance and energy efficiency capable of supporting emerging artificial intelligence and machine learning applications. Specifically, the three-year award beginning June 15, 2022 will explore a compiler approach allowing existing systems to use traditional data caches without hardware changes. It will also design an energy harvesting cache combining benefits of write-back and write-through caches. Finally, the project aims to introduce checkpointing for GPU registers and lightweight persistence for GPU shared memory in energy harvesting systems. This award is funded through the National Science Foundation's Computer and Information Science and Engineering program (CFDA #47.070), which supports investigator-initiated research and education across computing and information sciences.