Introduction to Cognitive Intelligence Computing Platform

The Cognitive Intelligence Computing Platform of Guangdong Institute of Intelligence Science and Technology (GDIIST) is established in 2021 and is an important part of the innovative organizational structure of the GDIIST. The Cognitive Intelligent Computing Platform is one of the three major technology platforms supporting neuromorphic and intelligent science and technology research.

With the technology and resource advantages of the GDIIST in neuromorphic computing architecture, multicore network-on-chip, neural network processing chip, wafer-scale integration and packaging, a professional and open resource sharing platform is established by uniting various researchers and technical personnel according to the discipline field and professional expertise.

The main task of the Cognitive Intelligence Computing Platform is to ensure the efficient operation of public scientific research facilities and equipment under a sharing mechanism and provide technical services, providing a development and verification platform for the cutting-edge research in the field of cognitive intelligence computing, facilitating the discipline development and cross-disciplinary collaboration in the GDIIST.  The Platform adopts the intelligent computing architecture of heterogeneous fusion of CPU/GPU/FPGA/BPU, and mainly focuses on the R&D needs of architecture, algorithm, chip and system in the field of advanced intelligent computing in the future. The Platform builds a complete set of prototype verification system that supports the research of neuromorphic chip, algorithm model development and trusted information link. The system mainly contains four dedicated development platforms.

1. Chip design and development platform: based on high-performance general-purpose CPU, it supports the common design and development process of chips, and provides an auxiliary design environment for hardware circuits for the trusted link development platform and the neuromorphic supercomputing prototype verification platform. This platform can cooperate with other development platforms to realize joint design and verification of software and hardware.

2. Algorithm model development platform: based on high-performance general-purpose GPU, it provides computing power support and a complete development environment for the development of traditional intelligence algorithms and cutting-edge neuromorphic intelligence algorithms, and provides real-time graphical computing capabilities for visual rendering. This platform can cooperate with other development platforms to realize joint design and verification of software and hardware.

3. Trusted link development platform: based on high-performance FPGA cloud accelerator, it supports the software and hardware development and verification of data center-oriented reconfigurable intelligent communication links, and provides blockchain-oriented SDK (Software Development Kits) and management platform. This platform can support the design of software and hardware collaborative computing systems for trusted links, and support the development and deployment of cloud-chain integration core technologies such as trusted computing environment, fast identity authentication, and hardware-assisted distributed consensus computing.

4. Neuromorphic supercomputing prototype verification platform: based on reconfigurable chips and integrating self-developed BPU (Brain Processing Units), it supports RTL design of ultra-large-scale neuromorphic chips, brain-inspired algorithm models, and distributed network-on-chip simulation test. The whole system can support the software and hardware co-verification of the neuromorphic chip architecture, circuits design and algorithm model of billions of neurons and hundreds of billions of synapses.

The system architecture and comprehensive performance indexes of the Cognitive Intelligence Computing Platform are generally in the domestic leading and international advanced level, with distinctive features in ultra-large-scale neuromorphic computing. It has important strategic significance and practical value for the future research layout and construction of the GDIIST in neuromorphic supercomputing, secure computing, metaverse and other cutting-edge information technology fields.