Reliable edge machine learning hardware for scientific applications

• DOI: 10.48550/arxiv.2406.19522
• Publication Date: 2024-06-27
• Authors: Tommaso Lisini Baldi; Javier Campos; Benjamin Hawks; J. Ngadiuba; Nhan Viet Tran; Daniel Díaz; J. Duarte; Ryan Kastner; Andres Meza; M. Quinnan; Olivia Weng; Caleb Geniesse; Amir Gholami; Michael W. Mahoney; Vladimir Lončar; Philip J. Harris; Joshua Agar; Shuyu Qin
• Keywords: FOS: Computer and information sciences; Computer Science - Machine Learning; Networking and Information Technology R&D (NITRD); Decent Work and Economic Growth; Information and Computing Sciences; Machine Learning and Artificial Intelligence; Software Engineering; Machine Learning (cs.LG)

Extreme data rate scientific experiments create massive amounts of that require efficient ML edge processing. This leads to unique validation challenges for VLSI implementations algorithms: enabling bit-accurate functional simulations performance in experimental software frameworks, verifying those models are robust under extreme quantization and pruning, ultra-fine-grained model inspection fault tolerance. We discuss approaches developing validating reliable algorithms at the such strict latency, resource, power, area requirements environments. study metrics algorithms, present preliminary results mitigation strategies, conclude with an outlook these future directions research towards longer-term goal autonomous experimentation methods accelerated discovery.