• Lead Pareto-optimizations on quality and efficiency for business-critical production ML models, including Gemini 1, 1.5, 2, 2.5 and Google Ads.

• Lead HW/SW co-design of TPUs from TPUv2 (the first generation of training TPUs), TPUv3, TPUv4/v4i, TPUv5e/v5p, TPUv6e, TPUs7 (Ironwood), and beyond.

• Build and lead a team of 40+ top-notch researchers and engineers from scratch.

• High performance and efficient cloud computing

• Deep Learning at scale (vision and language models on cloud) and edge (computer vision in autonomous driving)

Efficient Datacenter Servers and Memory Systems

Data Centric Computing and Manycore Systems