SiFive Launches New RISC-V AI IP with Scalar, Vector, and Matrix Compute - eeNews Europe

Abstract

SiFive has launched a new specialized RISC-V Intellectual Property (IP) core explicitly designed to accelerate Artificial Intelligence (AI) workloads. This new IP solution integrates comprehensive processing capabilities, combining traditional scalar processing with advanced vector extensions and dedicated matrix computation units. The product aims to provide a high-performance, programmable, and efficient platform for both edge and cloud AI applications using the open RISC-V architecture.

Report

SiFive Launches New RISC-V AI IP

Key Highlights

• AI Specialization: SiFive, a leading commercial RISC-V vendor, announced a new IP core specifically optimized for high-performance Artificial Intelligence (AI) and Machine Learning (ML) workloads.
• Unified Compute Architecture: The IP core features a integrated design combining three distinct compute paradigms: scalar, vector, and matrix computation.
• Market Target: This comprehensive approach positions the IP to compete directly in the rapidly expanding markets for AI inference and acceleration across various domains, including edge IoT and data centers.

Technical Details

• Scalar Compute: Provides the essential general-purpose processing capabilities required for control flow, system management, and running baseline operating systems or AI pipeline orchestration logic.
• Vector Compute: Integrates the RISC-V Vector (V) extension, enabling high-throughput Single Instruction, Multiple Data (SIMD) operations crucial for efficiently handling the linear algebra underlying most AI algorithms (e.g., convolution, pooling).
• Matrix Compute: Includes dedicated hardware acceleration units, typically specialized Matrix Multiplication (MMAC) blocks, offering maximum performance and efficiency for deep learning operations, significantly accelerating neural network inference tasks.
• Architecture: The core is based on the RISC-V Instruction Set Architecture (ISA), ensuring programmability and compatibility within the growing open-source ecosystem.

Implications

• RISC-V Maturity: The launch signifies the continued architectural maturation of the RISC-V ecosystem, demonstrating its ability to deliver specialized, high-performance IP that directly addresses domain-specific needs like AI, moving beyond general-purpose CPU cores.
• Competitive Pressure: SiFive directly challenges established architectures (like Arm and proprietary solutions) in the high-growth AI hardware acceleration market by offering an open, flexible, yet highly optimized solution.
• Increased Efficiency and Customization: By combining programmable vector units with dedicated matrix acceleration, SiFive provides system designers with the flexibility to handle emerging AI models while maximizing performance-per-watt for demanding applications.
• Broader Adoption: The availability of proven, commercially supported AI IP from a major vendor will accelerate the adoption of RISC-V into critical sectors requiring ML capabilities, such as automotive, robotics, and smart infrastructure.