uTensor is a framework that enables developers to deploy machine learning models on Microcontrollers (MCUs). AI inference today are typically done on high-performance computing units, such as, GPUs and TPUs. MCUs are low-energy and low-cost units. Deploying machine learning models on MCUs is a challenge because of their limited RAM size, flash capacity and clock-speed. uTensor addresses this by converting models directly into importable C++ source code, supporting quantization operators and using resource management schemes. In this talk, we will introduce Tensor, its run-time, code-generator and future work.
