EdgeCortix®Inc. has announced that it is delivering its cutting-edge Edge Artificial Intelligence (AI) Platform technology to its global Early Access Program (EAP) members. The company's flagship product is the EdgeCortix SAKURA-I AI co-processor, a high-performance, low-power, and energy-efficient turn-key AI co-processor that delivers up to 40 trillion operations per second (TOPs) of dedicated AI performance at sub-10 watts of power consumption. The SAKURA-I AI co-processor has been designed and engineered in Japan and is the company's first-generation semiconductor solution.
The EdgeCortix Early Access Program (EAP) is designed to offer customers the opportunity to assess EdgeCortix's products and services at scale by deploying them within their own complex, heterogeneous environments. The goal of the EAP is to showcase the ease of integration into customer's existing heterogeneous systems, enable customers to prove-out the effectiveness and efficiency of EdgeCortix solutions versus competing products, and facilitate a direct dialog with EdgeCortix product management, enabling tailor-made fit in certain cases.
The SAKURA-I AI co-processor is an essential component of EdgeCortix's Edge AI platform and is optimized for fast and power-efficient inference with one or multiple deep learning models at a time. It is particularly well-suited for real-time applications with streaming data, such as high-resolution video feeds, real-time controls for autonomous systems, radio-frequency signal processing, or 5G-Advanced and AI integrated applications. The SAKURA-I AI co-processor is implemented in TSMC 12nm FinFET, providing industry-leading inference performance into a small footprint in low power environments, with most models running on SAKURA-I consuming only an average of 2W-5W of power.
EdgeCortix's MERA software suite, which includes a heterogeneous platform compiler and toolset, has been available publicly for several months on Github. EAP members have previously been able to leverage production versions of EdgeCortix's software framework as well as its Dynamic Neural Accelerator (DNA), EdgeCortix's patented neural processing engine IP. The inclusion of the SAKURA-I AI co-processor into the EAP marks the start of a new phase in the development and deployment of EdgeCortix's Edge AI platform.
The SAKURA-I AI Co-processor is an artificial intelligence focused accelerator device that has been designed for rapid computations and power-efficient inference processing, from small to medium form-factor systems. It is especially well-suited for real-time applications with streaming data, such as high-resolution camera and lidar data, real-time controls for autonomous systems, radiofrequency (RF) signal processing, or 5G-AI integrated systems. The SAKURA-I co-processor delivers industry-leading AI inference capabilities across power and mobility-sensitive edge devices like aerial, underwater, or ground-based manned/autonomous platforms, smart city, smart manufacturing, visual and perception, and 5G systems.
EdgeCortix SAKURA-I has been benchmarked against leading-edge focused GPU platforms under different power modes. The Edge GPU is expected to be a TSMC 7nm device, while SAKURA-I is a TSMC 12nm device. End-to-end latency is measured under batch size 1, in all cases. All models deployed at INT8 without any pruning. All measurements on the Edge GPU were compiled and optimized by the vendor's latest tools. All SAKURA-I measurements were compiled and deployed on hardware using EdgeCortix MERA v1.3 software.
SAKURA-I Edge AI Processor Overview:
Up to 40 TOPS (single chip). Multiple chips can be combined on a single board.
Powered by EdgeCortix patented run-time reconfigurable data path - Dynamic Neural Accelerator IP.
On-chip dual 64-bit LPDDR4x – up to 16 GB on single board.
PCIe Gen 3 up to 16 GB/s bandwidth.
PCIe Device TDP at 10W-15W based on board form factor.
Dynamic power while running models on chip ~1W-5W.