With the continuous improvement of the level of social intelligence, the demand for artificial intelligence chips as the core of smart devices continues to rise. The favorable policies promote the vigorous development of the entire industry. A hundred flowers are blooming, and the road to localization of chips is advancing rapidly.
As a professional conversational artificial intelligence platform company in China, Spirit is constantly promoting the research and development and application of AI technology. In 2019, the chip company of Spirit, Shen Cong Intelligence, launched the AI chip Taihang TH1520, which is equipped with the full link of Spirit. The advantages of artificial intelligence voice technology and low-power algorithms make it widely used in smart home white appliances, black appliances and intelligent vehicle fields, promoting the scale and intelligent upgrade of traditional industries.
Recently, Shencong Intelligence has made another effort to release a new generation of artificial intelligence chip TH2608. This chip is the second generation of artificial intelligence SOC chip developed and built independently in the field of artificial intelligence after Taihang TH1520. slice and light up for verification.
Different from the TH1520 chip, the Taihang second-generation chip TH2608 is equipped with a Cortex-M CPU subsystem, a signal processor DSP subsystem, an efficient NPU subsystem, a voice and audio coding subsystem, and supports six analog microphones and six channels. Digital microphones and peripheral units enable the new generation of Taihang chips to quickly endow various products with voice interaction and device control capabilities, and at the same time upgrade the support for the Display control of smart products, and continue to expand the types and application scenarios of smart products. The ability of Taihang’s second-generation chips on the application side is more prominent. In terms of chip technology, the PPA (Power/Performance/Area) of TH2608 is further iterated and optimized.
In terms of voice capabilities, based on the full-link dialogue technology of Spiech, TH2608 also integrates command recognition capabilities, emotion recognition capabilities, voiceprint recognition capabilities and speech synthesis capabilities, which has been further optimized in terms of user experience. At the same time, it also increases the adaptability of multiple scenarios, such as multi-channel voice acquisition capability, rich and flexible interface configuration, Display capability, ultra-low power consumption wake-up capability, etc.
Compared with the first-generation chip, the second-generation chip has undergone greater functional upgrades:
First, the transition from TH1520 as coprocessor to TH2608 as the main control chip is realized. Built-in main control CPU based on Cortex-M ARM v8 instruction set, with abundant internal SRAM and external XIP Flash resources, as well as abundant and diverse peripheral interface resources, TH2608 can provide customers with main control processors in various application scenarios solution. At the same time, rich cache resources such as I-Cache/D-Cache and TCM are used, which greatly improves the processing efficiency and performance of the CPU. In addition, the floating-point computing capability of the CPU also provides additional flexibility for neural network operations and improves support for special networks.
Secondly, the highly flexible and configurable neural network processing unit NPU improves the energy efficiency ratio by a hundred times. Through the deployment of various general networks such as DNN/TDNN/FSMN/CNN, it can fully support data formats such as INT8/4/1, and with the help of the specially optimized network customization mode, the combination of complex networks with high efficiency and low energy consumption can be achieved. schedule. Compared with the use of conventional computing units such as DSP, the computing efficiency of neural networks has been improved by nearly a hundred times. Finally, it provides customers with abundant neural network algorithm support capabilities.
Third, provide a single-chip system integration solution, with a variety of flexible SIP multi-chip packaging forms. It supports the encapsulation of different types of PSRAM and Flash, and can integrate wireless communication modules such as WIFI and Bluetooth. At the same time, with the help of the built-in power management module (PMU), it has realized the strong support for the most streamlined solution of the customer’s BOM. The flexible configuration mechanism of IO pins and the fully customizable capability of external storage mode enable customers to achieve the feasibility of the most cost-effective configuration in various application scenarios.
In the application field, Taihang’s second-generation chip TH2608 continues to provide functions and A solution with comprehensive performance upgrade. Combined with the technical capabilities such as Voiceprint recognition, command recognition, and far-field interaction, TH2608 brings users a more accurate and flexible voice interaction experience. At the same time, continuing the achievements of Taihang’s first-generation chip as the first voice chip solution certified by Amazon, Taihang’s second-generation chip will continue to expand the implementation of product solutions in overseas markets.
The first-generation Taihang chip TH1520 was lit up and verified in 2019, and will be mass-produced and shipped in 2020. It will be applied in three major scenarios including smart home white appliances, black appliances, and smart vehicle fields. Midea, Hisense, Yadi, DingDipai and other companies have confirmed in-depth cooperation.
At the same time, Shencong Intelligent, a chip company of Spirit, completed the second round of financing in November 2020. In the same year, it was also selected into the “AI Chip Recommended Catalogue” of AIIA of the Ministry of Industry and Information Technology of the People’s Republic of China, and passed the international-level SGS three-system certification. In March 2021, Shencong Smart officially became the first voice chip company in the country to pass the Amazon Alexa certification in the United States.
At present, the evolution route of AI chips in multi-modality is clearer. The integration of voice, images, gestures and other interactive methods can greatly reduce the user’s AI experience threshold, which will also be the planning focus of Shencong’s next-generation chips. In the future, Shencong Intelligence will continue to strengthen the combination of algorithms and chips based on the advantages of Spirit’s full-link intelligent dialogue technology, give full play to its advantages, and continue to create products that can meet AI application scenarios.
The Links: NL2432HC17-01 FP75R12KT4-B15 IGBTS