AI chip companies leverage partners to target the Internet of Things market

Some time ago, because it was thought that the giants were heading for the next traffic-level entrance, the “smart speaker” was immediately sought after by the industry. Echo and Googleassistant were abroad, domestic cats and other youngsters launched their products, pushing the wave of this boom. On the top, 100 "boxes" were staged.

But after the war, it was the turn of the attention of the market to provide brain AI chips for smart speakers.

On October 31st, Hangzhou Guoxin, a veteran manufacturer of set-top box chips, held a conference to launch an AI chip GX8010 equipped with an NPU. The key application areas of the product will include smart speakers, voice interfaces, and preschool education for smart toys.

Ecology handed over to the giant build, AI chip wins in the specificity

Hangzhou Guoxin CEO Huang Zhijie

Dr. Huang Zhijie, the CEO of Hangzhou Guoxin, told the 36th that Guoxin’s goal was to transform from “digital life” to “intellectual life”. In early 2016, he created the artificial intelligence division and started to use AI chips. Technically speaking, some of the communication technologies, signal processing technologies, and video coding and decoding technologies used in the traditional set-top box chips are also useful in this AI chip, and there will be certain technical connectivity.

As deep learning has made major breakthroughs in various fields, it has imposed higher requirements on the computing power of processors. Therefore, building AI-specific chips is a concept for many companies. Previous CPUs can solve many linear problems, but they are not. Applicable to the era of artificial intelligence, dealing with deep neural network problems. The National Core is the same as most vendors. It wants to create special-purpose chips for the AI ​​in order to increase efficiency.

The newly released GX8010 has a built-in gxNPU neural network processor independently developed by Guoxin, which is called for artificial intelligence and is tailored to solve the problem of neural network acceleration and solving the inefficiency of traditional chips in neural network operations.

In addition, for the characteristics of small memory bandwidth in IoT devices, Guoxin specifically designed a neural network compression engine NCompressor to implement a one-button compression neural network. This can greatly reduce the memory capacity and bandwidth, while the speed of operation is improved.

What is the difference between this NPU and Google TPU, Huawei Unicorn 970 chip NPU? First of all, Google TPU is a processor that is applied to a server. It can be called cloud intelligence. It focuses on the size of the computing power and is not as sensitive to cost and power consumption. Compared with Google TPU, gxNPU is designed for the Internet of Things and aims at terminal intelligence. It adds a neural network compression engine, requires less memory and bandwidth for calculations, consumes less power, and is more suitable for various Internet of Things products. In deployment.

What is the difference between the "high performance and low power consumption" that terminal intelligence wants?

"Multi-core isomerism" and "multi-level wake-up" are the solutions provided by Guoxin for dynamic power consumption and standby problems.

National Core Product Manager RobotLing

Multi-core heterogeneous refers to the rational arrangement of the operating frequency and the timing of start and stop of each module. It can achieve the effect of using on demand and running out of service. For example, in a voice interaction scenario, the GX8010 needs 100-200 MHz to complete off-line speech recognition. The DSP works at 300-400 MHz to achieve multi-microphone array processing. The CPU can also dynamically adjust the system load. This solution allows the chip to operate efficiently while maintaining very low power consumption.

The standby problem has also been a difficult point for voice interactive devices. The system can still be awakened by voice when it is in standby, which means that a series of actions such as voice collection, noise reduction, and activation word recognition are required during standby. The multi-level wake-up mechanism of the GX8010 can perform hardware division based on whether there are sounds, whether there are voices, and whether it is a keyword.

In standby mode, the GX8010 uses the latest VAD (Voice Activity Detection) technology to detect whether the microphone has any voice input. Once the voice command is received, the DSP program starts noise reduction. The NPU activates word recognition. If the keyword is detected, the entire application system is activated. This step-by-step wake-up mechanism not only ensures that the voice assistant can respond to commands in real time, but also significantly extends the battery life of the device.

Leveraging partners to target the Internet of Things market

From the point of view of the conference of Guoxin, the number of cooperating companies present is a feature. From Ali, Tencent, as well as many other voice companies such as HKUST, Sibischi, Yunzhisheng, and outbound questions, all of them are on the list. In the conference hosted by the National Core, the guest speakers of Rokid and Si Bichi were also invited to give keynote speeches and participate in the post-meeting group visits.

Guoxin CEO Huang Zhijie said that the chip should be the result of multi-party cooperation. Why should there be so many partners? Rokid founder, Misa, also said at the scene: "Like we don't make chips ourselves, we don't need to have all the technology from top to bottom."

In terms of specific cooperation and division of labor, Guoxin will cooperate with Rokid in smart speakers in the future, and will work more closely with Sibischi in the smart home and automotive areas.

Huang Zhijie said that the goal of this chip is to achieve revenue in 2018.

The team of Hangzhou Guoxin is the team of Zhejiang University. Founder CEO Huang Zhijie graduated from Zhejiang University. RobotLing, the product manager of this chip, is the younger brother of Huang Zhijie at Zhejiang University.

In the 19th century California Gold Rush, the most profitable people were those who scouted gold? No, 99% of the gold diggers may not have digged gold, and those who benefit most are those who produce shovels. In this wave of smart speakers, smart home heat, the biggest benefit will be the production of "brain" AI chip business?

KNL4-100 Residual Current Circuit Breaker

KNL4-100 Moulded Case Circuit Breaker is MCCB , How to select good Molded Case Circuit Breaker suppliers? Korlen electric is your first choice. All moulded Case Circuit Breakers pass the CE.CB.SEMKO.SIRIM etc. Certificates.

Moulded Case Circuit Breaker /MCCB can be used to distribute electric power and protect power equipment against overload and short-current, and can change the circuit and start motor infrequently. The application of Moulded Case Circuit Breaker /MCCB is industrial.
Korlen electric also provide Miniature Circuit Breaker /MCB. Residual Current Circuit Breaker /RCCB. RCBO. Led light and so on .

KNL4-100 Molded Case Circuit Breaker,KNL4-100 Small Size Molded Case Circuit Breaker,KNL4-100 Electrical Molded Case Circuit Breaker,KNL4-100 Automatic Molded Case Circuit Breaker

Wenzhou Korlen Electric Appliances Co., Ltd. , https://www.korlen-electric.com