Alibaba pushes RISC-V processor for edge, AI, and autonomous driving

Alibaba’s chip division Pingtouge announced Xuantie 910 core processor IP last week to serve applications such as edge computing and autonomous driving.

The e-commerce giant intends to monetize the Xuantie 910 IP by licensing it to chipmakers. It is based on RISC-V, an open source chip architecture being developed by a number of companies. 

RISC-V is seen as a potential alternative to Arm architectures, which cannot be sold to Huawei and several other Chinese companies under a U.S. ban on national security grounds. 

China’s government is also urging its domestic tech industry to beef up its chip technologies as part of a larger goal of becoming less dependent on companies abroad.

Alibaba, an e-commerce giant in Asia, set up the Pintouge subsidiary last September with the acquisition of C-Sky Microsystems. The direction of the new processor is headed more in the direction of task-specific processors that serve the fragmented needs of AI and IoT rather than processors that serve universal needs like those used in PCs.

The Xuantie 910 is a 16-core RISC-V chip that runs on a 12nm node. Pingtouge claims it is 40% more powerful than any prior RISC-V processor.

What may be equally significant about the chip, however, is that Alibaba’s unit is making the product instead of relying on a more traditional chipmaking source like Intel. Large companies that are traditionally known for e-commerce or for writing software or running cloud environments like Facebook, Amazon, Apple, Netflix, Google, Tencent, Baidu and Microsoft are increasingly expected to join Alibaba in designing their own chips for AI.

Apple already purchases A-series chips from TSMC, the world’s largest foundry, based on its own designs.

The trend of these hyperscale companies designing their own chips is “likely to transform the semiconductor industry and put more pressure on traditional players in semiconductors,” wrote Alan Patterson for EE Times in a blog.

One central role of these AI-purpose chips is to reduce power consumption in data centers. While traditional chipmaker Intel will unveil its AI-focused Nervana NNP-L1000 in late 2019, it is not expected to match the performance of recent data center GPUs from Nvidia, he said.

RELATED: "The changing world of edge computing"