Perhaps this is the embedded AI processor that the Internet of Things needs

Recently, French semiconductor startup GreenWaves Technologies (GreenWaves) announced that it has received a total of 7 million euros in Series A financing, led by Huami Technology, Soitec, the company's angel round lead investor, and other investors. The funds will help the company continue to focus on providing disruptive ultra-low power embedded AI processors for edge computing devices that rely on batteries for long-term operation.

Huang Wang, founder, chairman and CEO of Huami Technology, said, "GAP8 and future products developed by GreenWaves Technologies are benchmarks in IoT low-power, high-performance signal processing and machine learning algorithms, and have broad application prospects in the field of IoT, including wearable devices. At the same time, this investment will strengthen Huami Technology's position and technology reserves in the RISC-V ecosystem."

As we all know, current semiconductor manufacturers are betting on the IoT market. The emergence of this unique startup may have a fierce impact on the "AI+MCU" IoT processor market that existing semiconductor manufacturers are promoting. GreenWaves also stated that the funds raised this time will be used for the sales of the company's first chip GAP8 and the research and development of the company's next-generation products. A market battle may be staged.

The game changer of embedded AI processors

In recent years, with the rise of artificial intelligence and the Internet of Things, the market has been chasing after edge-side artificial intelligence. Embedding AI in security equipment, smart watches or any other devices has become a hot spot pursued by manufacturers. However, considering the performance issues of existing solutions, a large number of companies have emerged in the market to develop their own chips for edge-side applications, and GreenWaves Technologies is one of the leaders.

GreenWaves was co-founded in 2014 by Loic Lietar, former executive vice president and chief strategy officer of STMicroelectronics, and other industry experts. In the past few years, the company has been committed to designing disruptive ultra-low power embedded solutions for AI processing in sensor devices such as image, sound and vibration. In February 2018, the company finally successfully released its first generation product GAP8 manufactured using a 55nm ultra-low power process.

GAP 8 Processor Architecture

As the world's first IoT application processor, GAP8 is designed with eight plus one RISC-V-based high-efficiency cores and embedded extended instruction sets, which gives it an extremely energy-efficient architecture. Secondly, its hardware convolution acceleration engine can speed up neural convolution calculations by three times and save energy by five times. In addition, it integrates power management and other functions. GreenWaves has set six working modes for the entire SoC, so that the voltage and frequency can be dynamically adjusted accordingly, and the power consumption is only the power consumption required for any given task. It is revealed that in deep sleep, the power consumption is only 1uA, and even under full load, the power consumption is only about 80mw. This energy efficiency ratio is not available in most processors on the market.

Loic Lietar also emphasized that GreenWaves has always insisted on the optimal energy efficiency ratio, with the goal of enabling AI to run for a long time on battery-powered edge devices, which is also the direction the company has been practicing and working towards.

GAP 8’s unique differentiated positioning

Official data shows that GAP8 is 20 times more energy efficient than existing products on the market in applications of understanding and analyzing information (images, sounds, etc.); the overall system cost is two to three times lower than other solutions. That means GAP8 can enable AI applications in various fields, and even when using batteries, this processor can keep these AI applications running normally for many years. GreenWaves compared GA8 and STM32F7 at different frequencies for CNN-based image recognition (Cifar-10), and it can be seen that the former has a great advantage in energy efficiency.

In May 2018, the company launched GAP8-SDK and GAPuino development boards, and sold more than 300 sets. According to GreenWaves, these customers are applying the solutions to various fields such as image and voice. Specific application scenarios include package recognition, product classification, pedestrian detection and sound monitoring. As for vibration, Greenwave believes that their processors will bring customers equipment maintenance efficiency and cost improvements in the industrial field. They have even developed a number of reference designs for different applications to simplify the design of potential customers.

GAP8 - SDK and GAPuino development board

Loic Lietar told the reporter of Semiconductor Industry Observer that GAP8 will be put into mass production in the second quarter of 2019 and many collaborations are also underway.

No fear of challenges,

The next generation of products will be greatly improved

As mentioned at the beginning of the article, GreenWaves will not be the only one interested in this trillion-dollar IoT market. Many manufacturers including Arm, ST, NXP and Renesas are also eyeing the direction of "AI+MCU". For example, STMicroelectronics has released its dedicated convolutional neural network accelerator code-named Orlando Project for ultra-low power MCUs; Renesas has a configurable coprocessor; even Arm has launched an extended instruction set for this market. With so many giants around, how can GreenWaves break through?

Loic Lietar said that the fact that so many manufacturers are paying attention to this market, especially the decision of Arm to promote the extended instruction set, proves the feasibility and correctness of GreenWaves' business in many aspects. In terms of technical reserves, they started relevant research and development three years ago. Even if there are new entrants now, it will take two years to see the effect. At this time, GreenWaves' first-mover advantage is prominent.

At the same time, GreenWaves' new generation of products will be unveiled this year, and the outstanding performance improvement will further expand GreenWaves' advantage. It is revealed that this product will use the best process in the field of low power consumption, which will better control the power consumption of the new chip; at the same time, the product's memory capacity will be greatly improved, and it will be able to support higher resolution images and more complex machine learning models. "Compared with GAP8, the new generation of products will be more energy efficient," Loic Lietar emphasized. Let's wait and see!

By Li Shoupeng, Semiconductor Industry Observer

*This article was originally created by the public account Semiconductor Industry Observation (ID: icbank). If you need to reprint, please add WeChat ID: icbank_kf01, or reply to the keyword "reprint" in the background of the public account, thank you.

Today is the 1865th issue of content shared by "Semiconductor Industry Observer" for you, welcome to follow.

Moore Elite is a leading chip design accelerator with the vision of "making it easy to make chips in China". Its business includes "chip design services, supply chain management, talent services, and enterprise services", and its customers cover 1,500 chip companies and 500,000 engineers. We are committed to providing ASIC design and Turnkey solutions, from Spec/FPGA/algorithms to chip delivery, including: chip architecture planning, IP selection, front-end design, DFT, verification, physical design, layout, tape-out, packaging and testing services. Since 2012, our team has been focusing on accumulating technical capabilities to help customers achieve optimal chip performance, and support flexible service models such as Turnkey, NRE, professional consulting and on-site. Moore Elite currently has 230 employees worldwide, headquartered in Shanghai, with branches in Beijing, Shenzhen, Hefei, Chongqing, Suzhou, Guangzhou, Chengdu, Xi'an, Nanjing, Xiamen, Hsinchu and Silicon Valley.