SiFive Expands Its RISC-V Intelligence Family to Address Exploding AI Workloads (www.forbes.com)

SiFive has expanded its RISC-V Intelligence Family with five new processor IP cores, targeting a broad spectrum of AI workloads from ultra-low-power edge IoT devices to high-performance AI data center tasks. Highlighting the launch are two all-new designs—the X160 Gen 2 optimized for energy-efficient edge compute, and the X180 Gen 2 aimed at more demanding edge inference and data center integration. Upgraded versions of existing cores (X280, X390, and XM Gen 2) complement the lineup, delivering enhanced memory subsystems with configurable caches and improved latency tolerance, as well as expanded datatype support including BF16 for AI training and vector crypto extensions. These vector-enabled RISC-V cores stand out by processing data in parallel to reduce instruction overhead and power consumption, which is critical for AI models using smaller datatypes. New interfaces—SSCI and VCIX—enable these processors to serve as Accelerator Control Units, facilitating low-latency, high-bandwidth integration with custom AI accelerators via RISC-V custom instructions. Performance gains are notable; for example, the X160 Gen 2 doubles the MLPerf Tiny benchmark scores over rivals with similar footprints, suiting it to constrained environments like wearables and smart home devices. Meanwhile, the XM Gen 2 scales to thousands of TOPS, addressing intensive generative AI workloads in data centers. SiFive’s announcement is timely amid rapid growth in AI workloads, especially at the edge, and signals a push to challenge incumbents like Arm and NVIDIA by combining the open, customizable RISC-V architecture with efficient scalar/vector compute and flexible accelerator control. Integration with major AI runtimes such as TensorFlow Lite and ONNX Runtime further positions SiFive’s solutions as adaptable and cost-effective. Success will depend on software ecosystem alignment and production scalability expected by 2026, but these advancements could significantly expand RISC-V’s role across AI domains.