Notifications
Back

Deep Learning 

Chip Design Solution

Deep Learning Chip Design Solution for You

Make your own AI chip for large-volume applications

We provide artificial intelligence chip designs for cases where large quantities of artificial intelligence chips are required. 

A high-speed, low-power AI design can be mounted on an AI board with a shape optimized for your use, and it can be applied to infinitely large quantities at no additional cost.

Applied chip

  • Xilinx FPGA

Features

  • It is a design whose stability has been verified by commercializing it in Deep Runner AI set-top devices for many years.
  •  It supports deep learning classification and object recognition algorithms with the highest recognition rates.
  •  It is the only design in the world that can recognize high-speed video on a low-end chip like the Zynq 7020. Thus, a low-cost board solution can be implemented.
  •  In general, AI chips execute AI algorithms with integer arithmetic calculations. Integer type calculation experienced a slight decrease in recognition rate in applications that require recognition of subtle differences, such as quality inspection. Our design runs the AI algorithm using floating point calculations, so there is no recognition rate degradation problem.
※ The 16-bit floating-point computation result completely overlaps the 64-bit floating-point result (not visible in the graph).
※ The 16-bit floating-point computation result completely overlaps the 64-bit floating-point result (not visible in the graph).
  • This design allows AI algorithms to run at high speed even with as little as 3.5 watts of power.
  • It supports minimal latency delay, making it suitable for applications requiring agile response.
  •  Consulting service and various schematics are provided together for the case of directly designing the AI board.
  •  Instead of designing an AI board yourself, you can use a variety of existing Zynq boards.
  •  You can use TensorFlow for model training, and we provide Windows-based Deep Trainer training software developed by us.
  •  We provide deep learning model training services upon request.

Applications

  • AI Education
  • AI Kiosk
  • AI Retail store
  • AI Factory
  • AI Inventory
  • AI Auto checkout
  • …

Supported deep learning models and computational speed

Recognition result examples

< EfficientDet D0 >
< EfficientDet D0 >

Supported Deep Learning Model Recognition Rate Comparison Table

Board configuration examples