Toshiba Develops DNN Hardware IP for Image Recognition AI Processor Visconti™5 for Automotive Driver Assistance Systems

January 7, 2019

Toshiba Electronic Devices & Storage Corporation

TOKYO—Toshiba Electronic Devices & Storage Corporation (“Toshiba”) today announced the development of Deep Neural Network (DNN [1]) hardware IP[2] that will help to realize advanced driver assistance systems (ADAS) and autonomous driving functions. The company will integrate the DNN hardware IP with conventional image processing technology and start sample shipments of Visconti™5, the next generation of Toshiba’s image-recognition processor, in September 2019.

The DNN hardware IP draws on deep learning to deliver more accurate detection and identification of a wider range of objects than image recognition based on conventional pattern recognition and machine learning. It enables Visconti™5 to recognize road traffic signs and road situations at high speed with low power consumption.

Toshiba will promote Visconti™5 equipped with DNN hardware IP as a key component of next-generation advanced driver assistance systems.

Advanced driver assistance systems such as autonomous emergency braking are now being widely adopted, from luxury cars to subcompacts. They are also expected to offer increasingly advanced capabilities - for instance, the 2020 version of the influential European New Car Assessment Programme (Euro NCAP), the EU-backed safety standard, adds testing to avoid collisions at intersections. This trend will increase the demand for more advanced and capable systems.

Toshiba Electronic Devices & Storage Corporation will continue to develop the Visconti™ family and contribute to traffic safety.

Outline of Visconti™5

Product Series Name

TMPV770 series

CPU core

Arm® Cortex®-A53

Arm® Cortex®-R4 processor with Floating Point Unit

Image processing DSP

General DSP

Image Processing Accelerator

Affine conversion

Pyramid Image Generator

Enhanced CoHOG Feature-based Support Vector Machine

Dense Optical Flow

Template Matching

Dense Stereo Matching

Deep Neural Network

Image Signal Processor

Video input interface

MIPI CSI-2 RX

Video output interface

MIPI CSI-2 TX

Notes

[1] DNN is a machine model using artificial deep neural networks inspired by human brain function.

[2] IP (Intellectual Property) is a function block for SoC development.

* Visconti™ is a trademark of Toshiba Electronic Device & Storage Corporation

* Arm and Cortex are registered trademarks of Arm Limited (or one of its subsidiaries) in the United States or other countries.

* MIPI is a registered trademark of MIPI Alliance.

* All other company names, product names and service names may be trademarks of their respective companies.

Customer Inquiries:

System LSI Marketing Dept.III
Tel: +81-44-548-2422

Information in this document, including product prices and specifications, content of services and contact information, is current on the date of the announcement but is subject to change without prior notice.

開啟新視窗