@article{endtoendtransformeraccelerationthroughpr,
  title = {End-to-End Transformer Acceleration Through Processing-in-Memory Architectures},
  author = {Xiaoxuan Yang and Peilin Chen and Tergel Molom-Ochir and Yiran Chen},
  year = {2025},
  eprint = {2601.14260},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2601.14260},
}