SEL's Approach to Artificial Intelligence (AI)


September 2017
Semiconductor Energy Laboratory Co., Ltd.



Semiconductor Energy Laboratory Co., Ltd. (SEL) is researching applications of the cutting-edge oxide semiconductor technology to artificial intelligence (AI), which has attracted attention in recent years.
Though AI has become explosively popular and is being actively pursued, power consumption of AI is a great challenge.
Our oxide semiconductor technology can significantly reduce power consumption and increase the efficiency of arithmetic operation. AI prevalently utilizes an artificial neural network, which is modeled after the human brain. Utilizing the OSFET® in hardware-implemented artificial neural network makes analog arithmetic possible, which allows us to create AI chips that consume drastically less power than conventional chips for use in edge devices.

For more details, please visit the following pages.
http://www.sel.co.jp/en/technology/os_ai.html
http://www.sel.co.jp/en/news/news/2017_05_ai.html

We are planning to exhibit our product at some exhibitions, to spread this technology to the world as well as Japan. One of the exhibitions is AI WORLD CONFERENCE & EXPO, held on December 11 to 13 in Boston, USA.
https://aiworld.com/

Our technology was exhibited at 1st AI EXPO held on June 28 to 30, 2017 in Tokyo Big Sight, Tokyo, Japan and received acclaim. Here are some of our exhibits.

Anomaly Detection with AI for Lithium-ion Batteries
~to create a safe control system for lithium-ion secondary batteries~


We propose a control system of secondary batteries with AI.
AI learns charging characteristics and predicts future charging characteristics when the battery is charged. Comparing prediction data and measurement data, AI detects anomalies to control the operation. Thus, battery problems can be prevented.

Image Sensor Integrating osAI~to improve the accuracy of recognition of objects with the image sensor~

Conceptual Diagram of Image Sensor with Embedded AI


Arithmetic operations of neural networks for AI can be made drastically efficient with our crystalline oxide semiconductor technologies.

Layer Structure of AI-integrated Image Sensor


This demonstration shows an image sensor with an embedded image processing block, enabling AI processing in the sensor chip. With use of AI, the image sensor can recognize humans, obstacles, traffic signs, etc.



The demonstrator shows an example of using an image captured by the image sensor for a side view mirror of the car. The display is flexible and is prototyped at SEL.

*Updated on September 15, 2017