A few days ago, at the 2021 Songshan Lake China IC Innovation Summit Forum, a total of ten local companies introduced new products and new technologies. It is worth noting that at the Songshan Lake Forum, the products introduced each time are basically mass-produced or about to be mass-produced. The products of , not only exist on PPT, so it is very suitable for engineers who need domestic substitution or localization to view.
Jingwei Qili Wang Haili: EDA tools are the biggest differentiation of FPGA
Wang Haili, CEO of Jingwei Qili, introduced the company’s current situation and product roadmap.
Wang Haili said that the four core elements of FPGA include: architecture design technology, IC design technology, EDA development technology, and IP development technology.
For FPGA products, Jingwei Qili has a wealth of product subdivision categories, including mountain series HME-M 65/55nm, river series HME-R 40nm, Hercules series HME-H 40nm, Pegasus series HME-P 40/22nm . Among them, the highest-end 22nm Pegasus P series is comparable to Xilinx’s V6. It adopts 6-input look-up table technology and has a number of patented technologies.
Wang Haili emphasized that if Jingwei Qili started from its predecessor, Jingwei Yager, it began to invest in tool development in 2005, including a complete set of software development tools from synthesis to layout and routing, FUXI, which is also the key to domestic FPGAs wanting to gain a foothold. .
At present, Jingwei Qili has formulated two three-year plans. One is from 2017 to 2020, focusing on the layout of small and medium-sized products and breaking through the consumer market. There are already H1 and H2 products in mass production, with shipments reaching 10 million pieces. . The second three-year plan is from 2020 to 2022, with the layout of mid-to-high-end product series P0, P1, P2, and P6, focusing on breakthroughs in communication, power, security, video and other customers. The two three-year plans effectively combine the company’s short-term urgent tasks with medium- and long-term development goals to ensure the company’s healthy, stable and sustainable development and achieve the final strategic goal.
Pengkan Technology: PonCan technology or solution to industrial bus interconnection technology
Li Chunchao, founder and CEO of Pengkan Technology (Shanghai) Co., Ltd., introduced a new generation of PonCAN industrial control network chips: VN1110/VN1810, which are mainly used in robots, intelligent manufacturing, and electric vehicles.
Pengkan Technology is a semiconductor design company dedicated to the design, research and development of innovative high-speed industrial control area network technology, providing a new generation of data transmission technology and complete semiconductor solutions for industrial control networks for robots, electric vehicles, and Industry 4.0.
The most important part of Pengkan Technology is PonCAN technology, which is also the origin of the company’s Chinese name Pengkan. PonCAN technology combines the characteristics of Pon (optical fiber) and CAN (industrial field control bus), which not only conforms to the real-time nature of CAN, but also highlights the It has the characteristics of high-speed transmission and low cost of optical fiber. According to Li Chunchao, PonCAN is a new-generation bus architecture that integrates high-speed data transmission, ultra-low latency and high-reliability control functions. PonCAN network has high security, easy wiring and synchronization of the whole network. Pengkan Technology’s first-generation product, VS1101/VS1801, provides high-performance PonCAN network and supports high-precision industrial control and edge computing.
For example, on the intelligent flexible joint (SCA) controller, a new control system chip is urgently needed, which needs to integrate control and network communication, reduce and reduce wiring, resist electromagnetic interference, reduce system complexity, simplify software system development, and have high precision, high The requirements of performance and high power can be perfectly satisfied by the Pengkan SoC.
In recent years, the field of autonomous driving has developed rapidly, and it has also brought many communication requirements. The total bandwidth requirement of sensor data may reach 3Gb/s-40Gb/s. Pengkan Network can empower in-vehicle networks and promote the evolution of in-vehicle domain architecture. Li Chunchao pointed out that the domain architecture mode of automobiles cannot completely match the data transmission requirements, so the industry has also proposed Zonal architecture (regional architecture).
In smart manufacturing, Industry 4.0 also requires the establishment of a strong and secure network to achieve electronics, digitization and analysis. PonCAN can meet its requirements for extremely high connectivity. At the same time, in terms of system cost, PonCAN also has great advantages over Ethernet, and customers can also support optical fiber or traditional copper network according to their needs.
In the industrial network architecture, PonCAN has the characteristics of high speed, low latency, high reliability, high-end security, and simplified connection.
At present, Pengkan’s business model mainly includes three types of PonCAN products, namely core network, bridge and node. Core network products are only sold in the form of modules, and these products manage the entire PonCAN network. Bridging products are sold only in modules, which provide bridging of different network domains. Node products are sold in chips, with SDK. Manufacturers can integrate proprietary software and control algorithms for greater flexibility.
Aixin Technology: Edge AI Vision Processor
Qiu Xiaoxin, founder and CEO of Aixin Technology, introduced the company’s high-performance, low-power artificial intelligence vision processor chip: AX630A, which can be widely used in smart cities, smart retail, smart communities, smart homes, and IoT devices. and many other fields.
Qiu Xiaoxin said that the current demand for computing power on the side is increasing, including multi-spectral, multi-sensor perception fusion, multi-modal intelligent perception, multi-scene interactive perception, low-light full-color, structured recognition, multi-algorithm fusion, etc. The types of algorithms are becoming more and more The more it is, the more complex the end-side algorithm is, and the more demanding the end-side computing power is. Due to limitations such as network latency, data bandwidth, etc., AI processing on the end-side becomes more important.
As a young enterprise, Aixin Technology, in response to market demand, from scratch, deeply integrates algorithms and hardware, continuously designs and develops high-performance artificial intelligence chips, and expands its influence in the industry. Qiu Xiaoxin elaborated on the four characteristics of AiCore Technology: “AiCore Technology’s AI ISP technology breaks through the bottleneck of traditional ISP, enhances image quality through iterative algorithm models, and makes images clearer; mixed-precision deep learning technology accelerates through proprietary networks. Unit, the perfect integration of chip algorithms; memory-computing collaborative design, reducing DDR read and writing, saving memory bandwidth; intelligent video encoding, efficient video compression, saving video storage.”
As a mass-produced high-end edge-side and end-side AI chip, the AX630A consumes about 3W, has a computing power of 57.6TOPS @800MHz 2w4f, and an equivalent computing power of 28.8TOPS @800MHz INT4. The number of images processed per second exceeds more than Thousands. Compared with its competitors, it has achieved a tenfold improvement.
Shiqing Technology: Using RISC-V for End-to-End Artificial Intelligence Vision
Yu Xin, President of Shiqing Intelligent Technology, brought the end-to-side smart vision chip AT5055 with RISC-V architecture. The chip will be officially released in July this year and is oriented to the fields of smart vision, smart home, and smart city. Shiqing Technology builds heterogeneous end-side artificial intelligence chips around RISC-V, DSP processors, and DSA processors.
AT5055 is a RISC-V architecture, high-performance, high-efficiency, low-cost end-side intelligent vision chip. The AT5055 is equipped with the TM800 main control processor and the TIMESFORMER intelligent processor of the RV32IMCF architecture developed by Time Engine Technology, which provides 0.6TOPS of high-efficiency artificial intelligence computing power, and also provides powerful DSP processing capabilities.
AT5055 supports a complete video channel for binocular vision, including high-performance ISP, H.265 video codec and 1080P Display solutions, supports 4+2 microphone arrays, and has built-in multi-mic noise reduction and local speech recognition algorithms.
AT5055 provides rich peripheral interfaces and a complete system security solution. Through the vertical integration of core processor IP, SOC architecture, algorithms and other core technologies, it has been deeply optimized for application scenarios such as face detection/recognition, social TV, smart IPC, etc., which can be empowered with the industry’s top energy efficiency ratio and cost performance. Device-side smart applications.
Yu Xin said that Shiqing Technology provides a one-stop RISC-V-based development environment, which does not require developers to pay too much attention to heterogeneous multi-core scheduling, thereby simplifying heterogeneous challenges and programming efficiency. Especially for AI algorithms, multi-party cooperation between processors, controllers and ISPs is not required, which reduces the communication barriers between different development teams of the company and fully ensures product performance.
Deep Cong Technology: Solve the voice interaction with more mouth and less hands
Wu Gengyuan, CEO of Shanghai Shencong Semiconductor, introduced the company’s low-power artificial intelligence human-machine voice interaction chip Taihang second-generation TH2608.
The second-generation chip TH2608 of Shencong Smart has been upgraded on the basis of the first-generation chip TH1520. 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.
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.
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.
Taihang first-generation products solved the problem of speech recognition from scratch, while Taihang second-generation products added AI local continuous speech recognition, semantic understanding and security features, voiceprint recognition and other functions, and used dedicated NPUs to improve energy efficiency and reduce power consumption. Shencong Semiconductor has many years of in-depth cooperation experience with SMIC and TSMC, and will continue to deepen the field of artificial intelligence voice interaction. Process, package fusion and optimization.
Relying on Spirit’s intelligent voice technology and resource empowerment, Shencong Intelligence focuses on creating an integrated overall solution of “algorithm + chip”. It has been established for three years and has shipped more than one million units so far. At present, the company focuses on applications such as smart home, smart car, wearable device, smart conference, and smart space.
Zhicun Technology: Wearable ultra-low power storage and computing integrated chip
Li Xiang, vice president of Zhicun Technology, introduced two revolutionary products of Zhicun Technology, including the world’s first integrated storage and computing accelerator WTM1001 and the first integrated storage and computing SoC chip WTM2101. WTM2101 is an ultra-low-power storage-computing integrated chip, which is used for low-power wake-up, recognition, noise reduction and other scenarios, and is used in TWS headsets, watches and other smart wearable devices.
The speed of data transfer is the bottleneck of computing. Faced with the flood of data, the problems of slow data transfer and high energy consumption need to be solved urgently. Zhicun’s integrated storage and computing technology innovation uses Flash memory to complete the storage and operation of neural networks, solves the problem of AI storage walls, improves computing efficiency and reduces costs.
The traditional Von Neumann architecture has some columns of data transfer at work, and the utilization of energy efficiency is very low, while the storage and computing integration does not have any data transfer process. Its essence is a vector-matrix multiplication. Conversion, convert the input data into an analog signal of current, when the signal enters the matrix, each unit in the matrix stores an 8bit number, such as storing 3, flowing through the transistor, the current is amplified by 3 times, and storing 100, the current Amplification of 100 times is equivalent to multiplication in the horizontal direction and accumulation of current in the vertical direction, so the flow through this matrix is equivalent to multiplication of a vector matrix.
The integrated technology of Flash memory and computing of Zhicun Technology includes three parts: Flash memory device (with several Flash memory pages), Flash memory page (with hundreds of millions of Flash memory units), and Flash memory units. Regarding why Zhicun Technology chose Flash, Li Xiang explained that, first of all, the Flash technology is mature, and it has been decades since the earliest mass production and use in equipment. Secondly, the storage density of Flash is high. Taking 40nm as an example, when the storage density is not large, the storage density of the Flash memory-computing integrated technology is several times higher than that of the 5nm SRAM. In addition, Zhicun Technology considers that the floating gate transistor used in Flash is a three-terminal MOS memory. Compared with other memories such as double-terminal MRAM and RRAM, the floating gate transistor is more similar to the MOS transistor, but it has one more floating gate than the MOS transistor. threshold voltage.
The size of WTM2101 is relatively small, 2.9*2.6mm, the peak computing power is 50Gops, the energy efficiency ratio reaches 15Tops/W, and the maximum neural network of 1.8M can be stored. In addition to the integrated storage and computing part, this chip also has a RISC-V core, audio ADC and power management, as well as rich interfaces. Due to its ultra-low power consumption, voiceprint recognition can be implemented on headphones and watches, bringing a richer experience to wearable devices.
Xinpu Technology: Simplifying PA for 5G Mobile Phone Design
Gu Jianzhong, COO of Xinpu Technology (Shanghai) Co., Ltd., introduced the company’s new 5G n77 1T2R RF front-end power amplifier chip XP5127. Compared with the first-generation 1T1R, the second-generation PA has a higher integration level, and the PA signal chain can be realized with four materials, while the first-generation requires seven.
The gain of XP5127 reaches 30dB, the transmit power can reach 29dBm, and the ACLR is -40dBc. In the PA module, because of the integration of chips of various different processes, thermal stress will also be generated inside when the ambient temperature changes. Therefore, PA modules often need to consider not only electrical performance, but also thermal performance, and the XP5127 also addresses this challenge in the design. The XP5127 adopts a fully differential scheme, which realizes heat dissipation through two-part differential PA, thereby improving linearity.
Gu Jianzhong said that according to the actual test results of the product, the performance is more advanced than the first and second foreign manufacturers.
The name of Xinpu Technology is derived from Simple, which is consistent with the company’s philosophy, which is to make more streamlined products and make it easier for customers to use.
Huajing Sensor: High Signal-to-Noise Ratio MEMS Microphone
Miao Jianmin, chairman of Huajing Sensing Technology, introduced its innovative product ML-2670-3525-DB1, the signal-to-noise ratio is greater than 70dB, and the acoustic overload point AOP is as high as 130dB.
The first core technology of Huajing Sensing Technology is the unique back plate technology. The current microphone back plate is mostly designed with acoustic holes of the same size, which is difficult to pass the strong air blow test of 0.8MPa. At the same time, the uniform aperture and hole spacing are not conducive to reducing acoustic noise and improving the signal-to-noise ratio. Huajing’s microphone back plate adopts a design with a large aperture in the middle and a small aperture in the edge, which improves the compression resistance, reduces acoustic noise, and further improves the signal-to-noise ratio of the microphone.
The second core technology is Huajing’s unique diaphragm technology. Huajing’s silicon microphone diaphragm adopts corrugated and fan-blade microstructures. Compared with the traditional flat plate structure, it improves sensitivity, consistency and yield. Eight fan blades automatically open to release airflow to avoid damage and fragmentation of the diaphragm. And this kind of diaphragm also makes Huajing’s products have better waterproof and dustproof characteristics.
Because the product has a high signal-to-noise ratio, it can achieve a high speech recognition rate at a distance of more than seven meters, which is very suitable for TVs, smart home appliances, etc. And a stronger signal-to-noise ratio can also be applied to TWS microphones to achieve better voice functions.
Minghao Sensing: Ultra-low-power three-axis accelerator
Wang Dawei, CEO of Suzhou Minghao Sensing Technology, introduced the company’s ultra-low power three-axis acceleration da7xx.
Today, consumer-grade MEMS sensor chips are a red sea. The MEMS industry has large investment, high risks, and long product development cycles. Wang Dawei said: “Minghao Sensing focuses on the development and innovation of MEMS sensing technology. We must say goodbye to ‘coolie’ and ‘low price’, continuously improve the ability of technological innovation and industrial chain integration, improve mass production execution, and in-depth business innovation. Mode change.”
Minghao Sensing has self-developed 3D MEMS-CMOS integrated micro-electromechanical process platform technology. The sensor is made of single crystal silicon with DRIE process and has a sealed silicon cap, which has higher sensitivity, strong output signal, and accurate output. Features, comparable to ST and Bosch products.
The device is packaged as a 2x2x0.98mm Land Grid Array (LGA). Compared with the BGA package, the LGA package has the technical advantages of miniaturization, high power density, and low noise of Electronic products, and is more suitable for application scenarios that require compact installation. The family features an integrated 32-level first-in, first-out (FIFO) buffer that allows the user to store data to limit host processor intervention. The da7xx family of sensors provides two independent and flexible interrupts that simplify the algorithms used to detect various motion states. Standard I2C and SPI interfaces are used to communicate with the chip, and are widely used in VAD voice wake-up assistance functions, precise voice positioning, call noise reduction assistance, bone voiceprint recognition assistance and other fields.
Space Intelligence: 5.8GHz microwave radar sensor
Zhang Zhenwei, Sales Director of Gakusong, introduced the company’s new products, including the radar sensor chip AT5820, the industry’s first mass-produced 5.8GHz microwave radar chip AT5810S, and the world’s first uA microwave radar chip AT5815.
AT5820 is the world’s first 5.8GHz microwave radar sensor SoC that can detect breathing. Compared with millimeter-wave radar, it has stronger penetration ability, longer transmission distance and lower price.
AT5820 integrates high-performance radar transceiver and 32-bit MCU on a single chip. It has rich resources and powerful performance. It can be interconnected with the main control or transmission chip, and can also be used as the main control MCU. It is a high-performance, high-sensitivity radar sensor. Due to the extremely high sensitivity of the product, it can detect the breathing and even the heartbeat of a living body, thereby realizing human detection. It can be used in many fields such as smart home, home appliances, lighting and so on.
Zhang Zhenwei said that from 2019 to 2021, airborne radar chip shipments will double, and monthly shipments in June will reach 3.5kk, with radar sensors accounting for 30% of the market share. Zhang Zhenwei predicts that by the end of the year, the shipment volume of airborne radar will exceed 5KK in a single month, becoming the largest radar manufacturer in terms of shipment volume. At the same time, Space Intelligence is also developing UWB and other products, thus becoming the manufacturer with the longest radar product line in the world.
“Industrial-grade quality, consumer-grade price, making all things easier to perceive and connect more intelligently.” This is the company vision of Airspace Intelligence. To this end, the company has cooperated with more than 20 solution providers such as Tuya Smart and Kuzhai, serving more than 1,000 customers. It is precisely by building an extensive ecological chain that the company’s products can be rapidly expanded.
As Wei Shaojun, chairman of the IC Design Branch of the China Semiconductor Industry Association, pointed out in his speech, the Songshan Lake Forum has three main characteristics: distinctive, grounded and close to the market. If any semiconductor company wants to succeed, it must keep abreast of market applications and meet market needs.
Therefore, we see that whether it is a company that has been established for more than ten years or a start-up company just two years old, it is a combination of unique products to conduct product or industry discussions. This is also the purpose of the Songshan Lake Forum.