The Current State of RISC-V Architecture for Edge AI Workloads

It’s still early days for RISC-V chipsets in Artificial Intelligence (AI) workloads, with just 1.8 million shipments anticipated for 2024. However, ABI Research forecasts significant market growth for the open Instruction Set Architecture (ISA), with 129 million annual shipments expected by 2030. RISC-V’s flexibility in addressing various workloads and its scalability are highly appealing factors for companies aiming to support edge AI applications

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

Log in or register to unlock this Research Highlight.

RISC-V Market Overview

It’s still early days for RISC-V chipsets in edge Artificial Intelligence (AI) workloads, with just 1.8 million shipments anticipated for 2024. However, ABI Research forecasts significant market growth for the open Instruction Set Architecture (ISA), with 129 million annual shipments expected by 2030. RISC-V’s flexibility in addressing various workloads and its scalability are highly appealing factors for companies aiming to support edge AI applications.

A major growth driver for RISC-V chips is edge AI gateways, as RISC-V can address more demanding AI inferencing workloads such as computer vision in automotive applications and security use cases. Our analysts also assert that personal/work devices (Personal Computers (PCs), laptops, smartphones, tablets), robotics, and wearables are other high-value areas for RISC-V chipsets.

The RISC-V opportunity is not going unnoticed. A multitude of companies—semiconductors/System-on-Chip (SoC) suppliers, device manufacturers, and software vendors—are actively embracing it. RISC-V International and the RISC-V Software Ecosystem (RISE) project are gaining significant traction, with the former experiencing a 24% increase in membership in 2023 alone. Notable players like SiFive, Venatna, Axelera AI, and Tenstorrent, as well as industry giants such as Google, MediaTek, NVIDIA, Intel, Qualcomm, and NXP Semiconductors, are all deeply committed to RISC-V development. Their engagement in working groups like RISC-V International and the RISE project is a testament to the potential of RISC-V in igniting innovation in the industry.

“ABI Research has seen a groundswell of activity around chipsets containing the RISC-V architecture specifically targeting AI workloads at the edge. The amount of startup activity in RISC-V addressing edge AI is also noteworthy, and several companies, such as Tenstorrent, have ambitious roadmaps with significant funding and partnerships to boot.” – Paul Schell, Industry Analyst at ABI Research

 

Background of RISC-V Architecture

RISC-V, with its modularity and scalability, emerges as a competitive option against established proprietary architectures such as Arm and x86. Its flexibility, particularly in AI and Machine Learning (ML) applications, combined with the cost-saving advantages over licensed alternatives, has driven global interest in the technology.

SiFive, founded in 2015 by researchers from the University of California Berkeley, leads in licensing processor cores based on RISC-V. Other companies like Axelera AI, Ventana, and Untether AI have also capitalized on the growing recognition and open-source enhancements of the ISA, including vector extensions for accelerating AI workloads.

Further Reading: Intel Foundry Update—A Plan to Secure Western AI Chips

RISC-V Breaks Historical Barriers to Efficient Edge AI

Untether AI and Axelera AI tackle memory and computational challenges in edge AI by designing new compute architectures that bring memory closer to the compute node, reducing bandwidth bottlenecks. This integration of memory and compute enhances power efficiency and performance. RISC-V's advantage lies in its support for in-memory compute, enabling Central Processing Units (CPUs) and accelerators to be located on the same block, leading to efficiency and latency improvements.

RISC-V chips cater to a wide range of applications, from low-powered inferencing in Internet of Things (IoT) devices to demanding model training in on-premises servers. Ventana and Axelera AI showcase the ISA's suitability for on-premises AI inferencing applications like computer vision.

Chinese Companies Turn to RISC-V to Curb AI Restrictions

RISC-V has become a strategic technology in China amid increasing U.S. sanctions on the semiconductor market. As tensions rise in the AI arena, China is rapidly shifting toward domestic chipsets due to restricted access to Western hardware. To counter these restrictions, China is investing heavily in RISC-V, an open-source ISA not currently subject to sanctions.

However, this transnational collaboration has drawn attention, prompting some U.S. lawmakers to urge restrictions on China's access to RISC-V. In response to potential interventions, RISC-V International relocated its headquarters to Switzerland in 2020 to maintain neutrality.

U.S. measures targeting China's AI advancement affect the consumer and enterprise markets, including Graphics Processing Unit (GPU) export sanctions limiting access to high-performance semiconductors like NVIDIA's GPUs. Additionally, RISC-V provides an alternative for countries like Russia and China to diverge from U.S.-restricted semiconductor exports, with examples like Alibaba's deployment of RISC-V server clusters using indigenous chips.

Key Companies in the RISC-V Space

ABI Research has observed the following companies being active in RISC-V development:

Learn More about RISC-V

For a deeper overview of the most recent technology developments spurring growth for RISC-V in edge AI chipsets, download ABI Research’s RISC-V for Edge AI Applications report. This report is part of the company’s Edge AI Research Spotlight, which is a library of similar content ready for you to uncover.