A particular kind of AI deployment is the Edge AI, this is one of the biggest trends in chip technology.
Edge AI chips run AI processing on the network edge — or, in other words, on a local device rather the cloud or internal servers.
The benefits of edge AI are clear: running AI processing on a device itself, instead of in a remote server, offers big benefits to privacy and speed when it comes to handling these requests.
The edge and server chipsets are optimized for high performance and ultra-low power running sub one trillion operations per seconds (TOPs). Power efficiency is important for improved battery life on edge devices and lower power costs for server infrastructure.
Our Partner, OCE Technology, an Irish company based in Dublin, teamed up with Chinese manufacturer Orbita, to develop a new system-on-chip that combines advanced neural network and digital signal processing with multiple interfaces, radiation tolerance, and low power consumption.
Hisaor, from the Irish word for artificial intelligence, is a new generation of embedded Edge AI processor chips. These chips focus on front-end image processing and front-end signal processing allowing intelligent control. The chip is an ideal platform for acceleration of deep learning and neural network algorithms. The Hisaor chip is a heterogeneous multi-core architecture (CPU + AI accelerator), fabricated on a 22nm FD-SOI production process which has the characteristics of high performance, high reliability and low power consumption, with a good TID performance and Latch-up effect immunity.
These characteristics make the Hisaor an excellent AI chip for Space, aerospace, military, intelligent security, AIoT, intelligent manufacturing, intelligent transportation and all hi-rel application that requires AI processing directly on the device.
The Hisaor SoC is composed of multiple CPUs, multiple AI co-processor units (GPU+NNA), image processing unit, on-chip bus, peripheral interfaces, on-chip memory, etc.
Peripheral interfaces include:
- Gb Ethernet
- USB OTG
- Camera Link
- Rapid IO x2
- LVDS Display
- CAN x2
- SPI x2
- UART x6
- GPIO x 64
The first of this new generation of OCE Edge AI chips is the Hisaor 03A.
This unit has a floating-point processing capacity of 64 GFLOPs and a fixed-point processing capacity of 12 TOPs;
This unit is equipped with 4x ARM Cortex-A9 cores for as the main processing unit and an AI coprocessing subsystem with by 8x Graphic Processing Units (GPUs) and 8x Neural Network Accelerators (NNAs). This chip brings the possibility to interface with the most used connectivity standards.
The Hisaor03A SoC has 1MB RAM embedded on chip together with 32KB eFuse and 32KB BootRom. External Memory can be added using the dedicated interface for DDR3 / DDR4 / LPDDR2 / LPDDR3, NAND Flash, SPI NOR Flash.
OCE and Orbita are planning to enlarge the HISAOR family SoC to give to the customer the possibility of choice for a better fit for the application.
A big value added for the Hisaor SoC is the support to the standard AI software interfaces such as OpenCL\OpenVX, and can realize seamless docking with mainstream deep learning software frameworks such as TensorFlow and Caffe.
A development Kit will be soon available.
To know more about Hisaor Edge AI chips contact us.