An integrated circuit that is dedicated to training and executing neural networks, the architecture of AI software. The training side (deep learning) and execution side (inference engine) require massive amounts of calculations, and AI chips perform them in parallel many millions of times every second. See
neural network and
deep learning.
The Versal Chip
Along with other specialized processors, the top model of this Xilinx system-on-chip (SoC) includes 400 AI engines (see
Versal).
(Image courtesy of Xilinx, www.xilinx.com)