China AI chip unicorn Cambricon Technologies plans to launch MLU machine learning series processor chips for cloud-based applications within the next 18 months, with the chips to be fabricated with 16nm process of Taiwan Semiconductor Manufacturing Company (TSMC), the firm's co-founder and CEO Chen Tianshi has announced.
Announcing the plan at his firm's recent product presentation conference in Beijing, Chen said that Cambricon will debut MLU100 AI chips to support inference for application to datacenters and small- to medium-size servers, and MLU200 chips to support training for application to AI R&D centers of enterprises.
While these two AI chips will be manufactured with TSMC 16nm process, the company will not rule out using 7nm to 5nm process nodes for the fabrication of its other AI chips in the future, according to Chen.
Chen stressed that Cambricon will manage to occupy 30% of the China market for high-performance smart chips and embed one billion smart terminal devices worldwide with the firm's chips in the next three years.
At the conference, Chen introduced three new AI processor models: the Cambricon-1H8 for low-power consumption computer vision application; the higher-end Cambricon-1H16 for more general-purpose applications; and the Cambricon-1M autonomous driving applications.
Of the three, the Cambricon-1M has yet to hit the market, Chen said, adding that in terms of computing performance, this new processor is 10 times faster than its first AI processor Cambricon-1A designed especially for mobile devices.
Chen also revealed that Huawei's newly-launched Kirin 970 chip is equipped with the Cambricon-1A processor (neural processing unit, NPU). The Kirin 970 chip installed on Huawei Mate 10 has a HiAI mobile computing architecture comprised of four parts: CPU, GPU, ISP/DSP and NPU.
Article translated by Willis Ke
