<p>➀ The increasing complexity of AI models and the exponential growth of network numbers and types have led to a dilemma for chip manufacturers between fixed function acceleration and programmable accelerators;</p><p>➁ The general AI processing methods are not up to standard, and focusing on specific use cases or workloads can achieve greater power saving and better performance in smaller space;</p><p>➂ The trend of AI algorithm complexity is increasing, and the number of floating-point operations is increasing, and the trend line is only pointing upwards and to the right.</p>