Key Takeaways
• SiFive and Kinara have partnered to launch a USB-based enablement board offering bare-metal access to SiFive’s X280 RISC-V vector processors.
• The Xara X280 board combines SiFive’s 64-bit X280 cores with Kinara’s Ara-2 NPU, capable of up to 40 TOPS of AI performance.
• The board supports a range of AI workloads at the edge, including CNNs and transformer-based generative AI models like LLaVA and LLaMA.
• The solution is targeted at developers building custom silicon and edge inference systems requiring performance, efficiency, and transparency.
SiFive and Kinara Introduce USB-Based Board Offering Bare-Metal Access to RISC-V Vector Processors for Edge AI
A strategic move to bring real-time AI inferencing and vector processing closer to the developer
SiFive and Kinara have jointly announced the HiFive Xara X280 enablement board, a USB-based device designed to provide bare-metal access to SiFive’s RISC-V-based X280 vector processors.
The launch is aimed at giving developers and hardware designers direct access to test and evaluate the X280 IP in real-time edge AI use cases, without the overhead of operating systems or middleware.
The board integrates Kinara’s Ara-2 processor, which features two SiFive X280 64-bit RISC-V cores and Kinara’s proprietary NPU architecture optimized for AI inference.
“We designed the Xara board with the intention of allowing SiFive customers to evaluate the real-time behavior of the X280 IP before integrating it into their own custom chips,” said Jack Kang, SVP, WW Business Development, Sales, and Customer Experience at SiFive.
What’s Inside the Xara X280 Board?
The USB-powered device gives developers access to key capabilities for AI-centric edge deployments:
• Dual SiFive X280 64-bit vector cores for real-time processing, floating-point math, and tensor operations
• Kinara Ara-2 NPU with up to 40 TOPS performance for AI workloads at the edge
• Support for generative AI and transformer-based models like LLaVA and LLaMA
This makes the board particularly relevant for AI systems that require low-latency inference without reliance on cloud-based compute infrastructure.
Targeted for Edge AI and Custom Silicon Development
The Xara board enables developers to run performance-critical operations directly on silicon without the constraints of an operating system stack.
This opens the door for tighter system optimization, faster iteration, and more reliable benchmarking in early-stage development.
“SiFive’s RISC-V cores are a perfect complement to Kinara’s neural processor cores enabling performance- and cost-optimized end-to-end inference for edge AI use cases,” said Ravi Annavajjhala, CEO of Kinara.
With AI now being deployed in edge environments ranging from smart cities and autonomous vehicles to industrial robots and vision-based systems, the availability of such a board addresses the growing need for efficient, real-time, and cost-effective inference platforms.
Strategic Implications for RISC-V and Edge Computing
While SiFive is widely recognized for spearheading the adoption of RISC-V architecture, this collaboration with Kinara pushes its utility further into AI and edge markets.
The HiFive Xara X280 serves not just as a development tool, but also as a strategic evaluation platform that encourages chip designers to integrate SiFive’s vector-capable IP into custom SoCs.
Kinara, acquired by NXP in February 2025, continues to establish itself as a key player in AI acceleration at the edge, offering scalable NPU solutions that are increasingly critical for local, on-device computation.
• The board enables benchmarking of SiFive’s vector processing IP in isolation or with Kinara’s NPUs
• No operating system or driver stack required, reducing system latency and development time
• Provides early hands-on access to evaluate performance across edge AI use cases
The HiFive Xara X280 is more than just a dev board—it’s a signal of intent from both SiFive and Kinara. For SiFive, it’s about empowering developers to fully test and trust its vector-capable RISC-V cores.
For Kinara, it’s about showcasing its AI inference capabilities in real-world, power-sensitive environments.
As generative AI, transformer models, and multi-modal AI systems push deeper into the edge, developer-friendly hardware like the Xara X280 could be a cornerstone in the next generation of custom AI silicon.
For more news and insights, visit AI News on our website.
