Apr 28, 2017
DMP AI Processor IP “ZIA™ DV700” Enables AI at the Edge

DMP today announced the new DMP ZIA™ DV700, a high-performance, low-power AI processor IP for inference, which extends the DMP ZIA™ product portfolio further into the fast-growing deep learning application areas at the edge.

DMP AI processor IP, ZIA™ DV700, is designed specifically for low-power deep learning inference on the edge that enables to process a wide range data such as images, movies and audio. According to a study at Ernest & Young Institute Co., Ltd, the AI market is expected to grow to US$ 200 billion by 2020, as a result of the adoption of AI in various fields such as manufacturing, whole sale and retail, transportation and logistics, service industry, etc.

It is expected that collaborative distributed deep learning technology becomes very popular. “Learning process” which require high performance processing, is conducted on the cloud and “Inference process” which require real time processing, is conducted at the edge. By delivering real-time analysis at the edge, it is possible to solve challenges related to privacy and bandwidth, minimizing data volume required for transmission and discarding video data immediately after analysis. Bringing the power of deep learning to edge devices such as sensors and smart phones requires highly energy-efficient AI inference processor IP that can perform the large number of required operations while consuming extremely little energy.

DMP’s AI processor IP “ZIA™ DV700” is the fully programmable and configurable IP that can optimize performance and size, comprising DMP’s advanced processor technology and hardware optimization technology. ZIA™ DV700 deliver up to 5x better power efficiency than other GPU based AI processors to address high throughput applications such as surveillance camera, medical, office appliance, automotive, robot, drone and home appliance.

ZIA™ DV700 Hihlights

Hardware

• Small size: 2.3mm2@TSMC28HPC + running 750MHz
• High Performance & Low power consumption
  – GoogLeNet – 116 FPS@750MHz, 670mW
• Configurable IP core(convolution layer etc)
• CNN, RNN, LSTM, GoogLeNet support
• 8 bit weight compression support
• 32 bit RISC processor to control CNN
• FP16 support

Software(Application development environment)

• C level API
• Caffe / Tensor flow protocol buffer support
• 8 bit weight compression tool

Availability

Target Application

• Surveillance Camera
• Medical
• OA
• Car
• Robot
• Drone
• Home Appliance