Developers want to get the tools and hardware up and running quickly so they can evaluate and learn the new SiMa.ai edge ML platform. The Developer Kit provides an Out-of-Box set of all components that you need and a guide to make this a breeze and let developers focus on their evaluation.

The first step an ML developer will want to do is establish the ML model performance and accuracy on a target platform and assess the capabilities, time and effort in compiling models of interest. Palette™ provides quantization and compilation of a developer’s model using Palette’s compiler. Utilize Palette’s silicon software image build and deploy tools to program the developer board. The developer kit provides the ability to execute these builds and provide KPIs such as; frames per second (fps), latency, accuracy, % loading of compute resources and memory footprint. Palette, running in a Docker on a desktop, supports ARM cross compilation with libraries to generate code for the SoC, a platform build, test and deploy tool suite to create complete ML applications.

The next step an ML developer will want to do is integrate this model into a potential application or use case, including ML model pre and post processing functions. To accelerate the prototyping phase, Palette adds the ability to quickly code, build and evaluate a pipeline using your own ML models. Developers can Python scripting using SiMa APIs for the functional pipeline, avoiding the complex embedded optimization often needed for on-device execution. These APIs bind your Python code to the device execution environment and will execute on device as a complete functioning pipeline. Pushbutton build and deploy loads this image into the Developer board for execution. This proof-of-concept application can validate the functionality of the pipeline as well as provide an early demonstration vehicle.

Quickly bring a real-time data stream to the platform, execute a pipeline on this data stream and display the performance results in real-time running on the MLSoC silicon for use case demonstrations. GStreamer example pipelines included in the Palette software release can be run out of the box to demonstrate real-time streaming performance. Customer pipelines can leverage these pipeline designs that take advantage of GStreamer to get higher utilization of compute resources and higher fps. The developer board processes the pipelines and displays the metadata overlaid on the host PC. Add additional cameras to provide multi-camera processing capabilities.

The MLSoC Developer Board is based upon our PCIe half-height, half-length production board but is modified to operate on a lab bench environment with access to is a versatile board that uses the SiMa.ai Machine Learning System on Chip (MLSoC) device.

Some of the key features are:

• On-bench: PCIe edge connector covered in protective coating and a plate with four footings under the board, a micro- USB connector provides power for operation on a lab bench.
• In PC’s: The PCIe form factor (68.9mm x 160mm) using a standard 98-pin PCB edge connector to slot into any standard host PC or motherboard. A bracket is included to assist in securing to a PC.
• Low power board. Typical workloads 10-15W.  Supports PCIe Gen 4.0 up to x8 lanes, LPDDR4 x4, I2C x2, eMMC, µSD card, QSPI-8 x1, 1G Ethernet x2 ports via RJ45, UART x2, and GPIO interfaces.
• Machine learning accelerator (MLA) – providing up to 50 Tera Ops Per Second (50 TOPS) for neural network computation.
• Application processing unit (APU) – a cluster of four ARM Cortex-A65 dual threaded processors operating up to 1.15 GHz to deliver up to 15K Dhry stone MIPs, eliminates the need for an external CPU or PC host.
• Video encoder/decoder – supports the H.264/H.265 compression standards with support for baseline/main/high profiles, 4:2:0 sub sampling with 8-bit precision. The encoder supports rates up to 4Kp30, while the decoder supports up to 4Kp60.
• Computer vision unit (CVU) – consists of a four core Synopsys ARC EV74 video processor supporting up to 600 16-bit GOPS.
• Incorporated for optimized execution of computer vision algorithms used in ML pre and post processing and other pipeline processing functions.

On-Site Accelerate delivers a customized face-to face session with a SiMa.ai technical expert on the Developer Kit Bring-up and training on the Palette software to compile, build and deploy Python and GStreamer based pipelines on the Developer Board Incorporate your whole team to get up to speed quickly and be able to ask questions real-time as you walk through the development process.

• On-Site Accelerate – DevKit bring-up and training:
  MLSOC-DBBU-SW-Training-OA