Research Scientist, HW​/SW Co-Design; PhD

Our teams’ mission is to explore, develop and help product ionize high performance software & hardware technologies for AI at datacenter scale. We achieve this via concurrent design and optimization of many aspects of the system from models and runtime all the way to the AI hardware, optimizing across compute, network and storage. The team invests significantly into model optimization on existing accelerator systems and guiding the future of models and AI HW s drives improved performance, new model architectures and reduces cost of ownership for all key AI services at FB:
Recommendations and Generative AI.This is an exciting space that spans exploration and productization, coupled with close collaborations with industry, academia, Meta’s Infrastructure and Product groups. Collaborating closely with product teams, the team's mode of operation is going from ideation and rapid prototyping, all the way to assisting productization of high leverage ideas, working with many partner teams to bring learnings from prototype into  addition to the real-world impact on billions of users of the Meta products, our team members have won Best Paper Awards at prestigious conferences such as ISCA, ASPLOS, SOSP, and OSDI, with multiple papers selected for IEEE Micro Top Picks.

We regularly publish in ICML, NeurIPS, SC, HPCA, NSDI, VLDB, MLSys, and more. Overall, our work largely corresponds to the research communities of systems in general and especially systems for ML (MLSys, SOSP, OSDI, SIGCOMM, NSDI), hardware architecture (ISCA, ASPLOS), ML (NeurIPS, ICML, ICLR) and supercomputing (SC, IC…)

We are seeking a Research Scientist to join our AI hardware and systems team. You will focus on cutting‑edge research and development in low‑precision numerical formats and algorithms for accelerating both AI inference and training s role involves exploring the theoretical underpinnings and practical implementation of novel numerics, quantization schemes, and mixed‑precision techniques to push the boundaries of AI efficiency and performance across various hardware platforms.

• Conduct fundamental research into advanced low-precision and mixed-precision numerical formats (e.g., 8-bit, 4-bit integers, custom floating-point formats) for AI models
• Develop novel quantization algorithms, calibration techniques, and hardware-aware numerical strategies optimized for both training stability and inference efficiency
• Design and implement proof-of-concept solutions in popular AI frameworks (e.g., PyTorch) to validate research hypotheses and measure performance impact on state-of-the‑art models
• Collaborate closely with hardware architects, compiler engineers, and AI model researchers to co‑optimize numerical formats and algorithms for next‑generation AI accelerators
• Publish research results in recognized conferences (e.g., NeurIPS, ICML, ICLR, ASPLOS, ISCA, HPCA, MLSys, MICRO)

• Currently has, or is in the process of obtaining a Bachelor's degree in Computer Science, Computer Engineering, relevant technical field, or equivalent practical experience. Degree must be completed prior to joining Meta
• Currently has, or is in the process of obtaining, a PhD degree in Computer Science, Electrical Engineering, Applied Mathematics, or relevant technical field, or equivalent practical experience. Degree must be completed prior to joining Meta
• Research experience in one or more of the following areas: low‑precision numerics, quantization, computer arithmetic, mixed‑precision training/inference, or numerical optimization for AI
• Theoretical background and practical experience with AI models (e.g., CNNs, Transformers, LLMs, Diffusion Models)
• In‑depth experience of Python and experience with at least one major AI framework
• Must obtain work authorization in country of employment at the time of hire, and maintain ongoing work authorization during employment

• Experience designing and evaluating custom numerical formats for AI models
• Familiarity with low-level programming for specialized hardware (e.g., CUDA, HIP, Triton) or hardware description languages (HDL)
• Experi…