◆AI accelerator architecture with customized computing power: provide diversified specifications, and allocate the most appropriate hardware resources according to customer application specifications. ◆Neural Network Performance Analyzer with Accelerator Features: Provides immediate performance estimates, which are about 10％ worse than accurate hardware simulations, enough to allow users to optimize model design based on accelerator features. ◆Quantitative training process consulting and translation services: Provide a transparent QAT quantification process that can be debugged and analyzed, and is equipped with a hardware accelerator. ◆Although AI applications are already popular in our life, the development of dedicated AI chips is still faced with small quantities, many customized specifications, and lack of design standards, resulting in daunting challenges for chip companies to put into mass production. In particular, deep neural networks require huge computing and memory requirements, computing resources and power consumption make product market positioning difficult. A complete solution that can be customized, improves algorithm efficiency, and improves chip system efficiency is urgently needed to implement the AI chip design industry.

Technology

S11204I0037

Industrial Technology Research Institute(工業技術研究院)

Research institution

Trial production

2023/04/10

Patent_exclusive license,Patent_non-exclusive license,Joint development,Negotiate by self,

2025/04/13