Calling for high-quality projects in the semiconductor hardcore track! The Core Power Preliminary Competition has set off a new wave of enthusiasm

According to Jiwei.com, the preliminary round of the 2021 "Core Power" competition "Finding Good Chinese Projects and Gathering China's Core Power" came to a successful conclusion on June 3. Three comment guests and many professional institutional investors participated in this online roadshow throughout.

This core power event focuses on four popular tracks: high-performance AI inference chips, RF signal chain chips, 4D millimeter-wave radars, and high-performance interface IPs.

The guests who commented on this roadshow were: Intel Capital Investment Director Nian Yi Bo, Beyond Moore Fund Investment Director Guo Qihang and Gaorong Capital Executive Director Yuan Feng.

The first project came from Shenzhen Kunyun Information Technology Co., Ltd., and mainly introduced the two core advantages of Kunyun Technology.

First, the product has the world's leading computing power cost-effectiveness and higher chip utilization (up to 65%~95%), and the actual measured performance is higher at the same or even lower peak computing power.

The CAISA customized data flow architecture independently developed by Kunyun Technology is a high-performance AI computing architecture customized for deep learning neural networks. This architecture controls the order of calculations through the flow order of data flows, eliminates the extra time overhead caused by instruction operations, and enables the operator-level data flow graph of the CNN network to achieve efficient pipeline operations.

Based on the CAISA 3.0 architecture, Kunyun Technology has developed the world's first data flow chip with a chip utilization rate of up to 95.4%, a peak performance of 10.9 TOPS, an operating temperature of -40℃~125℃, and can be used in edge and cloud AI analysis.

Second, commercial implementation in vertical fields. Kunyun Technology has implemented AI applications around computing chips and generated commercial value. For example, Kunyun Technology's products have been applied in smart power, smart oil fields, chemical parks, smart construction sites, smart mines, smart cement and other fields.

The second project came from Xiaxin Microelectronics (Shanghai) Co., Ltd., which mainly introduced the technology, team and product advantages.

RF signal chain chips are high-frequency analog circuits, which have higher requirements than ordinary analog circuits and higher technical barriers. Xiaxin Microelectronics has four core technologies: the first is the RDPTM platform with independent intellectual property rights, which is reconfigurable RF technology; the second is the use of special PMU technology to greatly reduce power consumption; the third is the use of innovative near-zero intermediate frequency Near ZIFTM frequency conversion architecture to provide reliability; and the fourth is the use of subthreshold design technology in the device to successfully achieve mass production.

In terms of technology, Xiaxinwei uses agile reconfigurable RF technology as a breakthrough, greatly improving the advantages of RF chips in terms of sensitivity, linearity, power consumption and cost. It has many invention patents and sufficient experience in mass production of analog chips (the chips designed by team members have shipped more than 100 million units in total).

In terms of the team, the company has a technical team with an average experience of more than 20 years and rich technical reserves. The top scientific research team comes from well-known companies and universities in the industry such as Fudan University, Peking University, and the Chinese Academy of Sciences. All core R&D personnel have doctoral degrees.

In terms of products, Xiaxinwei's first chip has been taped out for mass production and has generated substantial orders. The company's first chip is an HPLC/RF dual-mode transceiver chip for the industrial and state grid markets. Compared with similar products from ADI and TI, it has better performance and a price reduction of 30%+, perfectly meeting the needs of customers such as the State Grid and the China Institute of Electrical Engineering.

The third project came from Shanghai Manku Technology Co., Ltd., which mainly introduced the domestically produced 4D imaging millimeter-wave radar solution and product advantages.

According to reports, Manku Technology's 4D imaging millimeter-wave radar solution is equipped with independent hardware accelerator HWA + domestic low-end processor MCU + domestic MMIC; it is equipped with 2-4 domestic MMIC cascaded to form up to 16*16 MIMO virtual channels; it is equipped with automotive-grade external DDR (memory) and does not use the on-chip SRAM, breaking the RAM space limitation and achieving ultra-large-scale MIMO signal cache processing; it is equipped with up to 5Mbps CANFD bus or vehicle-mounted Ethernet (compliant with 100BASE-T1 specifications), and outputs all point cloud information in real time for L4 domain controller fusion algorithm. This solution has truly achieved national localization.

Manku Technology's independently developed 4D imaging millimeter-wave radar product, Aries, is comparable to a 16-line laser radar and is the first choice for cost-effectiveness in L4-L5. Aries' 1.5MB memory can complete real-time signal processing; the vectorized radar signal processing algorithm achieves stable capture of high-dynamic scenes; the horizontal angular resolution is within 1° (laser radar-level detection capability); each RD unit can output 4-8 targets in different directions in real time; built-in 8000 points * 256 target-level ultra-large-scale online target tracking algorithm.

Manku Technology's 4D imaging millimeter-wave radar products have four major advantages: 1. Nationally produced design to ensure supply chain security after mass production; 2. High performance comparable to 16-line laser radar to support L4-L5 autonomous driving system requirements; 3. Product design according to mass production standards; 4. Product design according to automotive-grade standards.

The fourth project came from Niuxin Semiconductor (Shenzhen) Co., Ltd., which mainly introduced the importance of localizing IP cores and the company's product lines.

According to Niuxin Semiconductor, IP core refers to a verified, reusable design module, which is the most critical element and competitive embodiment of the chip design industry, and is the cornerstone of integrated circuit design. At the same time, the localization of chips has brought more rigid demands for the localization of IP. Vigorously developing the IP industry is an effective way to solve the localization of IC design. With the increasing demand for data centers, the importance of interface IP is self-evident. High-end interface IP is widely used in mobile terminals, data processing centers, and national defense fields. It has a high degree of technical concentration and can reflect the development level of the integrated circuit industry. The demand for localization is very urgent.

Niuxin Semiconductor has long been focusing on the development of high-end SerDes + DDR core interface IP. The company is mainly composed of R&D personnel with rich design experience. Most of them have more than 10 years of design experience, and the core team has been working stably for more than 9 years. The problem of "lack of chips and cores" has been solved through independent research and development of products.

The company's star products include Combo SerDes, FPGA Combo SerDes and LVDS. Among them, 28G 28G SerDes technology is a key interconnect technology in wired and wireless communication applications. Niuxin's 28G SerDes IP based on the domestic SMIC 14nm process was successfully silicon-verified in one tape-out, becoming the only domestic long-distance 28G SerDes IP in the industry.

The 2021 "Core Power" preliminary round ended perfectly on June 3. Next week is the final week of the "Core Power" preliminary round. We will arrange roadshows on June 10 and June 11 to meet you, so stay tuned!