SDICVC-invested Cambricon released a new generation of AI chip, the application ecosystem of “Terminal-Cloud Integration” begins to show results

On May 3, 2018, SDICVC-invested company, Beijing Zhongke Cambricon Technologies Co., Ltd. held a product launch conference in Shanghai and released China's first cloud-based smart chip, Cambricon MLU100 (and its circuit board products) and the third generations of IP product, Cambricon 1M. The release of the MLU100 chip not only marks the Cambricon as the first company in China having both terminal and cloud-based smart processors, but also marks the official initiation of the Cambricon’s “Terminal-Cloud Integration” product strategy, which will promote the company to quickly establish a downstream application ecosystem. 



The Cambricon MLU100 chip is mainly used for cloud-based machine learning and inference tasks and can support various types of cloud application scenarios such as visual, speech, and natural language processing. Its theoretically equivalent peak in balanced mode reaches 128 trillion fixed-point operations per second, the theoretically equivalent peak value in high-performance mode can reach 166.4 trillion fixed-point operations per second, and the typical board-level power consumption is only 80 watts and the peak power consumption does not exceed 110 watts. The MLU100 chip is a new milestone in the development of the Cambricon, marking the Cambricon as China's first company (and also one of a few companies in the world) with both terminal and cloud-based smart processors. MLU100 chip can not only complete various complex cloud intelligence tasks independently, but also can be perfectly adapted to the Cambricon’s 1A/1H/1M series terminal processors, allowing the terminal and the cloud to collaboratively process complex intelligent task on the basis of a unified intelligent ecology. 


The MLU100's board uses the PCIe interface, which continues adopting the excellent universality of the Cambricon’s products. It can support all kinds of deep learning and classic machine learning algorithms to fully meet the cloud intelligent processing needs with large data volume, multi-tasking, multi-modality, low latency, high-throughput. 



Cambricon 1M is the company's third-generation IP product. The 8-bit performance ratio of TSMC's 7nm process reaches 5 Tops/watt (5 trillion operations per watt), providing three types of processor cores (2 Tops/4 Tops /8Tops) to meet different levels of intelligent processing requirements in different application scenarios, and to further improve performance through multi-core interconnection. With the excellent completeness as its first and second generation IP products (1H/1A), the Cambricon 1M processor can support diverse deep learning models such as CNN, RNN, and SOM, and further support SVM. k-NN, k-Means, decision tree and other classic machine learning algorithms, support local training and provide a flexible and efficient computing platform for visual, speech, natural language processing and various classic machine learning tasks. It will be widely used in smart phones., smart voice box, smart cameras, smart driving and other different areas. At present, Cambricon’s terminal intelligent processor IP products have been applied to intelligent terminals with tens of millions-level computing capacity (including Huawei Mate10, P20 and Honor v10 and other four smartphones.). 



At the release conference, Lenovo, Sugon, and Iflytek, as partners of the Cambricon, also released products applying the Cambricon chip, which provided users with dozens of times of processing efficiency in a typical scenario. The industrial ecosystem shaped by the Cambricon has achieved initial success. 


The release of the MLU100 cloud chip officially marks the initiation of the Cambricon “Terminal-Cloud Integration” product strategy. IP licensing on terminal enables customers around the world to quickly design and produce chip products with artificial intelligence processing capabilities. Cloud-based smart chips can help global customers to be more efficient in processing cloud computing. The "Terminal-Cloud Integration" strategy of "Terminal-Cloud Integration" strategy of "Terminal IP licensing+ Cloud smart chip + Instruction set" can help companies quickly establish a downstream application ecosystem. 


Chen Tianshi, founder and CEO of Cambricon, said that the original purpose of the Cambricon was to enable the world to use smart processors. The Cambricon will uphold the spirit of openness and collaboration in the academic community and share the latest Cambricon technological achievements with its counterparts in the world in the form of processor IP licensing, enabling global customers to rapidly design and produce chip products with artificial intelligence processing capabilities. 


Background links


In August 2017, the National Science and Technology Fund of SDIC Venture Capital, joined by Alibaba Ventures, Lenovo Capital, CAS Investment,Zhongke Turing, Oriza Seed, completed the Cambricon’s A Round of financing. Driven by the investment, its series of processors has rapidly advanced its terminal and cloud-based product and marketization. The company quickly gained market recognition and grew into the world’s first unicorn company in the field of artificial intelligence chips. The company has increasingly played an irreplaceable key role in the artificial intelligence industry ecosystem and is expected to become a new major landmark achievement in this field. SDIC Ventures Capital always adheres to the principle of “Serving the country's innovation strategy and focusing on the transformation of scientific and technological achievements", continuous to play the leading role and explore and cultivate a group of leading independent innovation companies represented by the Cambricon, contributing to accelerating the implementation of innovation-driven development strategies and building a strong world of science and technology.