platoseedplatoseed
something new
something new and big
Baud Labs develops a 1-bit native silicon chip intended for AI workloads, aiming to enable more dense and efficient neural networks. It provides hardware-level innovation for training and inference with 1-bit neural network formats and integrates with PyTorch for software workflows.
The product is a 1-bit native silicon chip designed for AI workloads, covering training and inference with formats such as W1A4, W1.58A4, and W1.58A6. It uses a per-MAC architecture with components like a ternary MAC, mux, and adder to support 1-bit neural networks and higher density silicon. Software compatibility includes PyTorch, and there are FPGA demos and a full-chip simulated environment on GlobalFoundries 12nm process, with future tape-out plans for silicon deployment.
Who it’s for: AI/ML hardware researchers and engineers, semiconductor developers working on ML accelerators, and organizations exploring novel 1-bit neural network hardware for efficient inference and training.
Hiring/open roles; ongoing demos and simulations; roadmap toward first silicon tape-out
Previously Sarang was at Peloton and Caper (W16) (acquired) as their founding deep learning engineer for deploying large vision models on resource constrained qualcomm chips for use in vision based fitness equipment and smart shopping carts. He has multiple patents in computer vision and has shipped AI products at scale with thousands of DAUs. Sarang has bachelors + double masters in Physics from IIT Bombay and UC Berkeley. He loves playing music and is also a certified pilot.
Formerly “Neuralyze”, “Cerelyze”, “Zener”, “Front”, “SuperCraft”, “Ponder” · why startups rename →
Software and hardware to run neural networks faster and cheaper
AI for ASIC Design Verification
