◆SAKURA™-II AI Accelerator

EdgeCortix SAKURA-II is a state-of-the-art AI accelerator providing the industry's highest efficiency through the company's low-latency Dynamic Neural Accelerator (DNA). It is a low power consumption silicon device that delivers excellent performance with a small footprint, designed for applications requiring fast, real-time, batch size one AI inference.
SAKURA-II is designed to handle generative AI applications at the edge, considered to be the most challenging, allowing designers to create new content based on heterogeneous inputs like images, text, and sound. Within a power envelope of 8W (standard value), it supports models with billions of parameters such as Llama 2 or 3, Stable Diffusion, DETR, and ViT, meeting the extensive requirements of numerous edge-based generative AI applications in visual, language, audio, and other fields.

◆SAKURA-II AI Accelerator Module and Card

SAKURA-II M.2 module and PCIe card are designed to run the latest vision and generative AI models with industry-leading energy efficiency and low latency. Whether integrating AI capabilities into an existing system or starting a new design from scratch, these modules and cards provide a platform for AI system development. By utilizing the MERA software compiler framework, designers can use these platforms to quickly source pre-trained AI models, complete system designs, and perform post-training calibration and quantization.
These modules and cards can be directly inserted into existing M.2 or PCIe systems to instantly upgrade and add AI capabilities.

◆MERA Software Compiler Framework

MERA provides all necessary tools, APIs, code generators, and runtime for deploying pre-trained deep neural networks, serving as both a compiler and a software framework. MERA natively supports Hugging Face, PyTorch, TensorFlow Lite, and ONNX, offering software developers and data scientists a familiar workflow. MERA provides an entire software stack for developing edge AI inference applications from modeling to deployment.

◆Dynamic Neural Accelerator® Technology

Dynamic Neural Accelerator (DNA) is a flexible modular neural accelerator IP core equipped with a "runtime reconfigurable" interconnect between compute units, achieving outstanding parallelism and efficiency through dynamic grouping. EdgeCortix uses a patented approach to reconstruct the data path between DNA engines in real-time, achieving exceptional parallelism while reducing on-chip memory bandwidth. (Alternatively, by using IP, it enables faster and more efficient hardware execution.)
DNA IP integrates with the MERA software stack to optimize the computation sequence and resource allocation when scheduling neural network tasks.

Overview of Edgecortix

EdgecortixJapan

EdgeCortix is a fabless semiconductor company specializing in energy-efficient AI processing for generative AI workloads at the edge. Founded in 2019 with its R&D headquarters in Tokyo, the company offers a software-first approach through its patented "hardware and software co-exploration" system, which designs AI-specific reconfigurable accelerator processors from the ground up with an AI-specialized runtime. EdgeCortix's products support the advancement of applications in rapidly growing edge AI hardware fields such as telecommunications, defense, aerospace, smart cities, Industry 4.0, autonomous driving systems, and robotics.

• Manufacture website