Watch Out for “Made in China” Open Source Artificial Intelligence Chipsets

Subscribe To Download This Insight

By Lian Jye Su | 4Q 2018 | IN-5269

In September 2018, Xiaomi-backed Huami launched the world’s first AI wearable chipset based on RISC-V architecture. The launch of the chipset brings a lot of promise for open-source AI hardware in China. As one of the largest wearable companies in the world, Huami’s adoption of RISC-V chipsets means RISC-V ISA is ready for large-scale deployment. This is a significant boost for the open-source chipset project and opens immense opportunities for AI developers and implementors alike to develop purpose-built AI chipsets.

Registered users can unlock up to five pieces of premium content each month.

Log in or register to unlock this Insight.


World’s First Open-Source AI Wearable Chipset


In September 2018, Huami announced what the company claimed to be the world’s first AI wearable chipset. The Xiaomi-backed wearable company also claimed that the Huangshan No. 1is the world’s first wearable chipset that is based on RISC-V open-source Instruction Set Architecture (ISA). The chipset is likely to support on-device inference on smartwatches and fitness trackers manufactured by Huami.

This development follows the launch of CK092, an IoT Central Processing Unit (CPU) based on RISC-V architecture by C-Sky, an Alibaba-owned chipset company. During the launch in early September, C-Sky also announced a collaboration deal with Pinecore, a chipset company also backed by Xiaomi, to produce RISC-V chipsets for Xiaomi’s ecosystem.

What Is RISC-V, and Why Is It Important?


Until now, the advancement in AI has been driven by non-open-source computing hardware. As AI use cases start to proliferate, more and more developers are looking to create purpose-built AI hardware at an affordable cost. In the case of many purpose-built AI chipset developers, such as Movidius (since acquired by Intel), Graphcore, and Horizon Robotics, the cost of AI-specific chipset fabrication can cost up to millions of dollars, restricting the deployment of AI in use cases that are low in scalability, volume, and profit margin. This has led developers to search for other methods to develop purpose-built AI hardware that can deliver the desired performance without having to commit significant upfront cost.

The RISC-V Foundation is a non-profit organization focusing on the development of open ISA based on established Reduced Instruction Set Computing (RISC) principles. Fully open source in nature, RISC-Vallows users to develop purpose-built computing hardware environments based on its ISA. While the base user specifications of the ISA were released and frozen by the University of California Berkeley, multiple standard extensions are available for other operations. As such, RISC-V offers customizable and modular clean slate design, giving users the flexibility to develop application-specific chipsets for machine learning and deep learning frameworks without the need to pay any licensing fee.

The open-source nature of the chipset will also enable more industry players to develop their own high-performance and energy-efficient AI chipsets without having to pay large amounts of licensing or loyalty fees, sparking more innovation in the AI industry. By leveraging on RISC-V architecture, AI companies can take advantage of the standardization in architecture, such as memory model, and development processes like compiler tools, debugging, and tracing, to accelerate their chipset development processes. This is also the reason why RISC-V has drawn the support from many industry giants, such as Western Digital, NVIDIA, Microsoft, Qualcomm, and Samsung. RISC-V is already replacing many proprietary implementations of microcontrollers in the market and is currently moving into high-end processors, such as CPU. This is not to say that industrial giants like ARM and Intel will face fierce competition anytime soon because it takes decades for enterprises to build up technical expertise and know-how in chipset development.

Open-source projects like RISC-V, nonetheless, allow enterprises to collaborate with a community of developers in designing chipsets that fit their needs. Aside from the aforementioned C-Sky and Huami, startups such as GreenWaves Technologies and Esperanto Technologies have released AI chipsets based on RISC-V cores. All of these chipsets share three main features, namely purpose-built for AI, high performance, and minimal power consumption.

Win-Win Situation for Xiaomi and RISC-V


Development of the Made in China chipsethas been a focus of the government of China in recent years, as local technology companies are encouraged. Huawei and Baidu are among the first batch of companies that have developed their own chipsets. Leveraging RISC-V ISA allows Alibaba and Xiaomi to catch up with their respective competitors. Despite launching its first chipset, known as Surge S1 System-on-a-Chip (SoC) for smartphone in April 2017, Xiaomi has struggled to follow up with Surge S2 in 2018. While Pinecore continues to work on S2, Huami’s AI wearable chipset will ensure that Xiaomi can move forward with its plan to develop end-to-end capabilities in its IoT business. Xiaomi launched an open-source edge machine learning framework known as MICE this year, and its Huami chipset will allow the company to better integrate MICE into its edge AI chipset, to customize its AI solutions, and to design better IoT products.

On the other hand, RISC-V will also benefit from this partnership. Both Alibaba and Xiaomi boast large communities of developers and wide ecosystems and supply chains. This means support from OS vendors, cloud vendors, component vendors, and system integrators and implementers. In addition to a large user base, RISC-V can benefit from a large pool of contributors coming from different verticals and backgrounds. Given the multi-domain nature of AI, it is critical to have inputs from different verticals, particularly from the cybersecurity domain.

According to the ABI Research market dataon AI and machine learning, 1.2 billion devices capable of on-device AI inference will be shipped in 2023, with 70% of them coming from mobile devices and wearables. ABI Research believes that open-source hardware, championed by RISC-V, will bring forth a new generation of open-source chipsets designed for specific machine learning and deep learning applications at the edge. This will enable the proliferation of AI across various Internet of Things (IoT) domains as low-cost and purpose-built hardware is able to address vertical-specific requirements.