Imagination Technologies has announced that it has licensed the IMG Series3NX AI core to Hercules for integration into its latest programmable intelligent chip, Avatar FPGA. These programmable intelligent accelerator chips are expected to have wide-ranging applications that necessitate robust data computing power and logic programming. They are particularly suited for AIoT applications, as well as Edge AI-enabled video analytics and processing applications.
Hercules’ new accelerator chip integrates multiple heterogeneous computing units, including FPGAs, CPUs, and AI, onto a single chip. It adopts a domain-adaptive and logically reconfigurable computing model with features of “software programmable and hardware reconfigurable.” This novel single-chip architecture efficiently and flexibly addresses the intelligence requirements of diverse Edge devices. It enables AI-enabled analysis, discrimination, and prediction on massive on-device data, eliminating the limitations of traditional multi-type single computing chips in terms of computing power, power consumption, size, cost, and application adaptation. This advancement helps users to swiftly navigate from application concept design to the final mass production of products in emerging application areas.
Imagination and Hercules have jointly developed HME-A1MF676, an adaptive and reconfigurable heterogeneous computing platform chip. This chip seamlessly connects FPGAs with CPUs and dedicated AI (enabling some GPU functions like edge computing and inference) through an innovative network-on-chipThis hybrid platform chip offers superior customization and energy efficiency compared to single-architecture chips, with the combined functionality of the three unit types surpassing that of a single-function chip. The adaptive and reconfigurable computing architecture brings multiple benefits including effectively balancing the requirements of multiple applications, including low power consumption, low latency, high performance, high bandwidth, and diverse workloads.
“We are thrilled to collaborate with Imagination. As pioneers in China’s FPGA+X heterogeneous computing ecosystem, we have embraced the advancements in FPGA technology, enhancing the implementation efficiency of our products by effectively balancing computing power, programmability, power consumption, and cost. The integration of Imagination’s AI IP offers an optimal solution for our low-power, cost-effective intelligent chip products,” said Zhou Qingrui, Vice President of Product at Hercules
“This partnership marks the first of its kind between Imagination and Hercules, a prominent representative of the FPGA industry in China, aimed at accelerating innovation in heterogeneous computing. Our AI heterogeneous computing architecture and ecosystem boasts industry-leading power, performance, and area (PPA), along with exceptional flexibility, to cater to a wide range of applications in this field. We are thrilled about the expanded partnership, which will enable Hercules to extend its solutions to a broader spectrum of intelligent applications in the future, leveraging the scalability of our IPs.” Said James Liu, Vice President and General Manager of China at Imagination.