|
|
www.design-reuse-embedded.com |
SiFive Empowers AI at Scale with RISC-V Innovation
embedded.com, Dec. 04, 2024 –
Artificial intelligence is increasingly transforming industries, and adopting RISC-V as a flexible and scalable architecture plays a significant role in this shift. Ian Ferguson, senior director at SiFive, shared insights about the pivotal moments driving RISC-V adoption for AI, focusing on the value of flexibility, performance and scalability.
"AI is not just a standalone feature," he said. Instead, AI is being embedded across various applications, and SiFive's approach aims to ensure the integration of AI within industries such as automotive and data centers.
According to Ferguson, companies require flexible and scalable hardware solutions to maximize their AI software investments. The Intelligence XM Series is intended to cater to the varying needs of AI functionality–whether by accelerating CPUs using vector extensions or offering more capable AI offload engines.
AI integrated across applications
The XM Series emphasizes flexibility, critical for integrating AI into diverse implementations. Ferguson noted that the level of AI optimization can depend on the stability of AI algorithms. "How stable the AI algorithms are will determine the level of optimization versus the flexibility needed to support changes," he added. By addressing these considerations, SiFive aims to ensure that its solutions are adaptable to customer needs, regardless of complexity.
SiFive's technology is already being used by major players, such as Google, Meta and Nvidia. Ferguson said that RISC-V-based chips from SiFive are not limited to high-end data centers but span a wide range of applications, from connected microcontrollers to autonomous vehicles. "We envision AI as impacting the entire performance spectrum," he said, citing use cases such as voice-assistant wake words, image recognition for autonomous driving and recommender systems that filter optimized content for users.
While specific customers cannot be named, Ferguson confirmed that SiFive has over 400 design wins, with billions of chips already deployed. "Our customers are building platforms that harness products from our performance and intelligence families of processors," he said. This variety of use cases reflects the versatility and scalability of RISC-V for AI workloads...