Hon Hai and Huabang Electronics invested in an AI chip company.

Nong Neng Intelligence, which focuses on marginal AI, was founded in San Diego in the United States and now claims to have received investment from Hon Hai and Huabang Electronics.

Nai Neng and investors will cooperate closely in the field of AI, and the two sides will establish a strategic cooperation relationship. It is reported that Nai Neng will join the open platform of the MIH electric vehicle built by Honghai. MIH has regarded Nina as its partner at the beginning of its establishment, and this time it will work with Hon Hai to develop AI applications in the automotive industry. In addition, the cooperation between Nineng and Hon Hai will also jointly promote Industrial 4.0. Ning Neng and Huabang Electronics will devote themselves to the development of AI-based microcontroller (MCU) and memory computing (Memory Computing).

Earlier, Nina had received $40 million from Horizons Ventures, which is owned by Li Ka-shing. In addition, it has also been favored by global well-known capital, such as Alibaba Entrepreneur Fund, CDIB, Wonderland Optoelectronics, Qualcomm, China Science and Technology Chuangda, Weili, Sequoia Capital Fund Cloudatlas and so on, with a total investment of more than US $73 million.

Dr. Liu Juncheng (Albert Liu), founder and CEO of Nintendo, said: "at a time when many companies are cutting back on R & D, we thank investors for believing in the work that Nergy has done. Nineng is a young company that has made a lot of achievements since its establishment in 2015, and we strive for more achievements in the future, which can not be done without the support of many investors such as Hon Hai and Huabang Electronics. This is an exciting time to advance the development of AI chips, and endurance will continue to play a vital role in this field.

Endurance provides a complete end-to-end hardware and software solution that enables AI reasoning on the upper edge of devices in mobile devices, personal computers and IoT (including smart home devices, monitoring, payment and smart cars). The Endurance solution adds cloud-based AI, to accelerate AI reasoning on devices. It is understood that air conditioning giant Gree and autopilot software company Teraki are its cooperative customers.

Dr Liu Juncheng added: "We are very pleased to be able to move forward with our partners and investors. 2020 was an important year for Nanneng, when we released KL720, and joined senior generals such as Chen Junyu (Davis Chen), former general manager of engineering research and development in Taipei of Qualcomm. In addition to developing joint projects with our investors, more chips are planned to be released in 2021. Marginal AI is still a relatively new concept for many people, and we look forward to bringing this technology to everyone. "

With the rapid rise of marginal AI industry, its edge AI technology has been widely used in the market because of its early technology investment and commercial layout, which makes it a leader in the industry.

Endure a variety of products to help

It is understood that Enron now includes neural processing units (NPU), KL520 and KL720 and their solutions. Among them, Endurance NPU provides market-proven solutions to meet the needs of low-function, low-heat distribution and complex neural network computing.

KL520 and KL720 are two AI chips launched by the company. As previously reported, Kneron KL520 was built using the UMC 40nm process, with two Arm Cortex-M4 cores and a self-developed NPU. These two Arm cores play the roles of system control (ARM Cortex-M4@200MHz) and AI coprocessor (ARM Cortex-M4@250MHz) in SoC, respectively. The computing power of the whole chip can reach 345GOPS (300MHz), and the average power consumption is only 500mW. It can accelerate the neural network model from Endurance and third-party public devices, thus facilitating the realization of 2D / 3D visual recognition and audio recognition in daily equipment. It is suitable for 3D sensing technology such as structured light, ToF and binocular vision and calculates different neural network models, and has multiple advantages such as specification, performance and cost, and solves the problems of relatively expensive 3D module, high chip cost and high hardware power consumption.

Coming to KL720, is an ability to support 4K images, full HD (1080p) video and natural language audio processing, so that the device can capture more details for face and audio recognition. It is understood that in addition to integrating the KDP 720NPU developed by ourselves, it also integrates the function of Cadence's DSP, as a coprocessor. In addition, Mineng also integrates the Arm Cortex-M4 kernel for the new SoC to provide more control support for the design of the terminal.

Compared with the previous generation of KL520 chips, the frequency of KL720 NPU increases from 300Mhz to 700MHz, its peak speed in 8-bit mode also increases from 345GOPS and 576MAC/cycle of the previous generation to 1.5TOPS of this generation, and the frequency of the M4 core used for 1024 MAC/cycle; control is also increased from the previous generation of 200Mhz to 400MHz of this generation. Another point worth emphasizing about this new chip is that their AI coprocessor has been changed from the previous generation of ARM Cortex-M4 to this generation of DSP, because of the features of DSP, which will undoubtedly further increase the strength of their new SoC.

In addition to hardware, Enron provides edge AI algorithms that can be embedded in the industry's smallest memory based on the latest NIST test results, including facial detection, facial recognition, body detection and gesture recognition.

Reconfiguration is the highlight of its technology.

According to the previous introduction of Nanneng, the edge AI solution has the characteristics of reconfigurable technology. On KL520, they said that the company uses a reconfigurable chip design on this product, which increases the flexibility of the chip, which is why it can simultaneously meet the acceleration needs of a variety of image recognition applications, such as face, body, gesture, object, scene, car model, license plate and so on. Combined with its compression technology and dynamic storage resource allocation technology, the chip can further improve its resource utilization. Thus, according to the application needs of the device, it can switch between audio recognition and 2D / 3D visual recognition in real time. In addition, this reconfigurable technology is compatible not only with popular AI frameworks such as Tensor Flow,ONNX,Keras,Caffe and PyTorch, but also with major CNN models including VGG16,ResNet,GoogleNet,YOLO,Tiny YOLO,LeNet,MobileNet, Densenet.

Based on the reconfigurable technology features of Endurance, its vision for Edge AI Net and AIoT 3.0 can be realized. In short, Edge AI Net will democratize AI and create more Wall-E and EVA with less Skynet. Edge AI Net allows Edge AI devices to communicate with each other, creating a collective action platform that does not rely on centralized cloud AI services.

Tolerance relies on the perfect balance between performance and power consumption, memory footprint and low-cost solutions that make it a leader in the edge AI field. In addition, its performance is well above the "size" level of its model, which was strongly verified in tests by NIST facial recognition vendors in 2019, while its balance is critical in many cases, such as when storage space, size and power are limited, including everyday applications such as cameras, smart doorbells, smart door locks, smartphones, etc. In addition, the solution is compatible with the main AI platforms, and the real-time reconfigurable technology can adapt to different application requirements.