Alibaba Group, the biggest e-commerce company in China, is setting up its own chipmaking subsidiary, Pingtouge Semiconductor Company, to make its in-house artificial intelligence inference chips.
Jeff Zhang, Alibaba‘s CTO and the head of DAMO Academy, the company’s research arm, revealed the news at the Computing Conference in Hangzhou on September 19. The company’s name, which is picked by Jack Ma, co-founder and executive chairman of Alibaba, references a Chinese nickname for the honey badger.
Alibaba has previously made several investments into chipmakers, such as the purchase of Chinese microchip maker Hangzhou C-SKY Microsystems, the sole CPU IP Core producer in China this April. The new subsidiary, Pingtouge, is expected to combine the work of C-SKY Microsystems and DAMO Academy’s own chip research.
According to Jeff Zhang, Pingtouge is mainly to make customized artificial intelligence inference chips to support the company’s fast-growing cloud computing and internet of things (IoT) businesses. The subsidiary will also further the research of embedded processors, an area that C-SKY has been working on.
Pingtouge is also expected to develop quantum chips, one of the five focuses of the DAMO Academy. The company plans to launch its homegrown quantum chips in two to three years and promises to establish an internal laboratory aimed at quantum computing. The research into quantum chips is at present dominated by U.S. chipmakers such as IBM, Google, Intel and Microsoft.
Alibaba‘s aggressive dive into semiconductors comes as China’s government looks to raise the quality of home-made chips. Earlier this year, ZTE, a Chinese tech company and smartphone maker, was brought to its knees as U.S. chip suppliers were barred from selling their products to the company.
Alibaba has already developed its own chip, AliNPU, this April. According to Jeff Zhang, the AliNPU team has completed all the verification of chip indicators as of last week and the chip will be officially available in June next year. Zhang claimed that the deep learning reasoning performance has been improved by 10 times, and the performance and power consumption ratio is 40 times that of similar products.