Through AI acceleration, smart terminal applications are innovatively improved

Beijing, China - July 11, 2023 - Imagination Technologies announced that it has authorized Jingwei Qili to use the IMG Series3NX AI core in its latest programmable smart chip Avatar FPGA series . This type of programmable intelligent acceleration chip will be widely used in various applications that require both data computing power and logic programming, such as AIoT applications, edge AI video analysis and processing applications, etc.

Jingwei Qili’s new acceleration chip integrates multiple heterogeneous computing units such as FPGA, CPU, and AI on the same chip. It adopts a domain-adaptive and logically reconfigurable computing model and has the characteristics of “software programmability, hardware "Reconfigurable" feature. This type of new architecture single chip can efficiently and flexibly solve the intelligent needs of diverse edge devices, enable device terminals to perform artificial intelligence analysis, judgment and prediction on huge data, breaking through the computing power (performance) of traditional multi-type single computing chips. , power consumption, volume, cost and application adaptation constraints, helping users quickly complete the entire process from application concept design to final mass production of products in emerging application fields.

Through in-depth technical discussions with Jingwei, Imagination helps its FPGA+X heterogeneous computing ecosystem achieve computing power empowerment. Imagination and Jingwei jointly created a new adaptive and reconfigurable heterogeneous computing platform chip HME-A1MF676, which combines FPGA with CPU and dedicated AI (implementing some GPU functions, such as edge computing and reasoning) through an innovative on-chip network Interoperability, this hybrid architecture platform chip supports a higher level of customization and energy efficiency improvement than any single architecture chip. The combined force of the three types of units can far exceed the efficiency of a single-function chip by several times. The advantages of adaptive and reconfigurable computing architecture can effectively balance various application requirements such as "low energy consumption, low latency, high performance, high bandwidth, and multiple loads".