SiMa.ai’s MLSoC enables engineers to run ultra-efficient AI inferencing at the edge derived from their choice of machine learning model, network, and framework.
The MLSoC is a fully integrated solution that contains 4x Arm Cortex-A65 dual-threaded application processors, Synopsis’ EV74 computer vision processor, and SiMa-ai’s Machine Learning Accelerator so that all operations – including running Linux but excluding model training – can be performed in one place without the requirement for cloud connectivity. In addition to reducing BOM, AI integration is drastically simplified as bandwidth between discrete systems is no longer an issue.
SiMa.ai’s MLSoC is capable of up to 50 TOPS (Tera-Operations Per Second) at 10 TOPS/W, making it faster and more efficient than leading solutions from NVIDIA at the time of release. Key machine learning application areas include smart vision, industrial, robotics, drones, autonomous vehicles, healthcare, aerospace, and defence.
The architecture of the MLSoC is shown in the diagram below:
The MLSoC includes the following base-level interfaces, but the system can be expanded to support MIPI, CSL, GSML, and other camera interfaces for machine vision applications:
- 4x Gigabit Ethernet
- Up to 8 lanes PCIe 4.0
- 2x SPI8
- 2x I2C
- 32x GPIOs
- 4x LLPDDR4
- SDIO/eMMC
In addition to the previously outlined processing and machine learning capabilities, SiMa.ai’s MLSoC features dedicated hardware for decoding and encoding up to 8 and 4 HD video streams respectively, with native support for 4K video. This operates independently of the main CPU processing cores. Similarly, the MLSoC contains hardware security elements and supports secure boot operations.
SiMa.ai’s MLSoC is currently available in a 31 x 31-mm FCBGA package with variants for commercial- and industrial-grade temperature specifications. Additional qualifications and screening are available on request.
You can learn more about SiMa.ai’s Machine Learning System-on-Chip in this official product brief.
SiMa.ai provides several plug-and-play ways to integrate or evaluate its technology. This includes a PCIe compatible platform based on SiMa.ai’s HHHL Production Board for datacentre and industrial PC applications, as well as a dual-M.2 board for space-constrained embedded applications (pictured).
Fill out the form below if you have a commercial project where you would like to put the MLSoC to the test, and ipXchange will get you connected for further information and samples.
(Images sourced from SiMa.ai)