Towards artificial general intelligence with hybrid Tianjic chip architecture

Area of Science:

• Artificial Intelligence
• Neuroscience
• Computer Engineering

Background:

• Artificial general intelligence (AGI) development is hindered by incompatible platforms for computer-science and neuroscience approaches.
• A unified platform is needed to support diverse AGI models and algorithms.

Purpose of the Study:

• To present the Tianjic chip, a novel hybrid hardware platform for AGI.
• To demonstrate the chip's capability in integrating and processing diverse AI models.

Main Methods:

• Developed the Tianjic chip with a many-core architecture, reconfigurable building blocks, and hybrid coding schemes.
• Implemented both computer-science-based machine learning and neuroscience-inspired algorithms.
• Tested the chip in an unmanned bicycle system for simultaneous, real-time control tasks.

Main Results:

• The Tianjic chip successfully integrated and processed diverse algorithms and models.
• Demonstrated real-time object detection, tracking, voice control, obstacle avoidance, and balance control on a single chip.
• Showcased the chip's versatility in supporting hybrid AI approaches.

Conclusions:

• The Tianjic chip provides a generalized hardware platform that synergizes computer-science and neuroscience AGI approaches.
• This hybrid platform is expected to accelerate the development of artificial general intelligence.
• The demonstrated capabilities highlight the potential for more advanced and integrated AI systems.