China Resources Microelectronics (Hong Kong) Limited
The proposed solution aims to provide tiny devices with optimized machine learning architecture and algorithms to achieve low resource requirement, low power consumption and low latency. The tiny devices are micro controller unit based devices with hardware accelerators using approximate computing technology to implement convolution neural network. Our proposed solution will be used for voice recognition to perform automatic control for various applications such as electric automobiles and smart homes. The proposed solution enables IP vendors and ASIC companies to develop competitive smart devices. Our solution includes an innovative algorithm of approximate computing using multi-block floating point to reduce computation power and memory usage.
