K2 - C3: SoC with Vector Processor for Low Power Edge AI
Overview
This System-on-Chip (SoC) integrates a RISC-V core coupled with a Vector Processor, delivering a powerful yet energy-efficient platform for Edge AI applications. The RISC-V core provides flexible control and programmability, while the vector processor accelerates parallel data operations essential for AI workloads such as convolutional neural networks and real-time sensor analytics. This combination enables efficient handling of compute-intensive tasks like image recognition, voice processing, and anomaly detection directly at the edge, reducing latency and reliance on cloud resources. Designed for low-power environments, this SoC is ideal for smart cameras, industrial IoT, and portable AI-enabled devices.
Dive into the Details
01
RISC-V Core - RV32IMAC
The SoC features a 32-bit RISC-V RV32IMAC processor core that supports Multiplication (M), Atomic operations (A), and Compressed instructions (C) for optimized performance, code efficiency, and multi-threading. It includes bit manipulation support essential for cryptographic operations. With the ZiCSR extension, the core enables efficient access to Control and Status Registers (CSRs), providing critical support for RTOS features like privilege control, task switching, exception handling, and interrupt management in real-time systems.
02
Vector Processor based AI Accelerator
A high-performance Vector Processor-based AI Accelerator is integrated to handle parallel AI workloads efficiently. It processes entire data vectors in single instructions, significantly accelerating neural network tasks such as matrix multiplications and activation functions. Designed for data-level parallelism, this accelerator offers notable improvements in speed and energy efficiency over scalar processors, making it well-suited for edge AI applications including real-time vision, audio processing, and anomaly detection.
03
Software Development Kit (SDK) and IDE
To facilitate development, an SDK is provided for deploying Deep Neural Networks (DNNs) on the Vector Processor. It includes tools for classical crypto (AES/SHA) and post-quantum cryptographic (PQC) functions. Development is supported via the Eclipse IDE with integrated debugging tools, enabling efficient application creation, testing, and optimization for both AI and secure computing use cases on the SoC.
04
Real Time Operating System (RTOS) Support
The SoC supports RTOS integration to manage multiple tasks in edge AI applications such as sensor data acquisition, on-device processing, and result transmission. This is achieved through Control and Status Registers (CSRs), the ZiCSR extension, and a Core Local Interrupt Controller (CLINT). Both Zephyr and FreeRTOS are fully compatible, providing developers with flexible, real-time control over system behavior.