3 trends for 2024: AI drives more edge intelligence, RISC-V, & chiplets - embedded.com

Abstract

The embedded technology landscape for 2024 is being rapidly reshaped by the confluence of three major trends: the expansion of AI processing to the edge, the increasing ubiquity and maturity of the open-source RISC-V ISA, and the shift toward modular semiconductor architectures using chiplets. These developments collectively address the growing demand for highly specialized, efficient, and customizable computing solutions necessary for future embedded systems. RISC-V, in particular, benefits significantly from the push toward domain-specific hardware accelerated by AI demands and facilitated by chiplet integration.

Report

Structured Report: 3 Trends for 2024

Key Highlights

• Edge AI Acceleration: The primary driver for new hardware innovation is the necessity to move sophisticated Artificial Intelligence and Machine Learning (AI/ML) processing away from the cloud and directly onto embedded devices (the edge).
• RISC-V Maturation: The RISC-V Instruction Set Architecture (ISA) is predicted to achieve greater mainstream success, transitioning from niche applications into broader industrial and commercial deployments due to its flexibility and open nature.
• Rise of Chiplets: Modular semiconductor design, utilizing chiplets (small functional blocks integrated into a single package), is becoming the preferred strategy for building complex, customized System-on-Chips (SoCs) efficiently.
• Convergence for Specialization: These three trends converge to enable highly specialized, heterogeneous computing solutions optimized for specific tasks, particularly those requiring low-latency, high-efficiency AI processing.

Technical Details

• Heterogeneous Architecture: Edge AI requires integrating diverse processing units—general-purpose CPUs (often RISC-V based), specialized Neural Processing Units (NPUs), and Digital Signal Processors (DSPs)—to maximize throughput while minimizing power consumption.
• Custom Instruction Sets (RISC-V): RISC-V's open standard allows designers to add custom extensions and instructions tailored specifically to accelerate AI workloads or other domain-specific functions without licensing restrictions.
• UCIe and Interconnects: The adoption of chiplets relies heavily on standardized die-to-die interconnect technologies, such as the Universal Chiplet Interconnect Express (UCIe), to ensure interoperability and cost-effective integration of IP blocks from different vendors.
• Design Specialization: The focus shifts from high-performance general computation to power-efficient specialization, which is critical for battery-operated devices and industrial IoT applications.

Implications

• Democratization of Silicon Design: The combination of open-source RISC-V IP and the modularity of chiplets significantly lowers the barrier to entry for developing highly customized silicon, fostering innovation beyond traditional semiconductor giants.
• Growth Vector for RISC-V: The demand for edge intelligence acts as a massive catalyst for RISC-V adoption. Since AI acceleration requires domain-specific optimization, RISC-V's customizability becomes a crucial competitive advantage over fixed architectures.
• Supply Chain Flexibility: Chiplets introduce flexibility in the supply chain, allowing companies to mix and match best-in-class components (e.g., a proprietary AI accelerator chiplet with a commercial RISC-V CPU chiplet) and leverage varying fabrication processes.
• Future of Embedded Computing: This convergence drives the next generation of embedded systems, enabling capabilities like real-time computer vision and sophisticated sensor fusion directly on devices, fundamentally changing industrial automation, automotive technology, and smart consumer products.