NEDO: Edge AI chip: Power efficiency 10 times, SLAM processing time 1/20

NEDO: Edge AI chip: Power efficiency 10 times, SLAM processing time 1/20

NEDO:

June 18, 2020,

NEDO co-developed with Socionext, ArchiTek, and Toyota Industries.

“Hybrid Quantization Deep Neural Network (DNN) Technology”
“Evolutionary Virtual Engine Architecture Technology (aIPE)”
“Real-time SLAM processing technology”
Announced that it was developed.

Evolutionary, low power AI edge LSI:

We prototype-evaluated the “evolutionary, low power consumption AI edge LSI”.

Dramatically improved performance:

The power efficiency of AI recognition processing is more than 10 times that of general-purpose GPU,
Image processing power efficiency is 10 times more than general-purpose GPU,
Real-time SLAM time improved to 1/20 of CPU,
It is expected that low power consumption, low latency, low cost machine vision, security, monitoring, and in-vehicle sensing system will be built.

MONOist

https://monoist.atmarkit.co.jp/mn/spv/2007/03/news073.html

AI Edge LSI:

AI recognition/image processing efficiency 10 times,

SLAM time 1/20 achieved

Toyota Industries

https://www.toyota-shokki.co.jp/news/topics/2020/06/18/002672/index.html