![](https://d15shllkswkct0.cloudfront.net/wp-content/blogs.dir/1/files/2023/09/mlsoc-chip-1.png)
![](https://d15shllkswkct0.cloudfront.net/wp-content/blogs.dir/1/files/2023/09/mlsoc-chip-1.png)
SiMa Technologies Inc., developer of a purpose-built machine learning system-on-a-chip for edge devices, today announced the availability of its free visual development environment that enables developers to get started easily building artificial intelligence models that live at the network edge.
The startup offers a platform known as the SiMa.ai Palette, which provides a set of scalable tools for developing machine learning models with high performance and push-button operation. Its GStreamer workflows use advanced scripting and automation on optimized neural network models, and provide pre- and post-processing functions and pipelines that permit the deployment of applications to large numbers of edge devices simultaneously.
The new Palette Edgematic tool aims to simplify AI and machine learning development at the edge. It offers a no-code interface that can be used to create, evaluate and fine-tune applications through a secure browser.
According to SiMa, edge AI applications can be created using simple drag-and-drop functionality, with their performance and power consumption requirements evaluated in real time. Once the app is ready, it can be deployed and executed on a SiMa developer board or production board, the company said.
SiMa’s hardware, which was announced earlier this year, includes a PCI Express half-height, half-length plug-in board and dual M.2 production boards containing purpose-built silicon for low-power edge uses. M.2 is an AI-specific high-performance parallel computation machine. Customers can use the board designs to accelerate deployment and to develop their own custom devices.
The boards include four-core Arm A65 processors, an H.264-compatible video encoder/decoder, a machine learning accelerator that delivers up to 50 trillion operations per second of performance and a computer vision processor that can be used with all popular machine learning frameworks, including TensorFlow, PyTorch, ONNX and MXNe.
SiMa says its low-code development approach will bring about a “new normal” in edge AI and machine learning. It allows anyone with a computer vision pipeline concept to convert it into executable code and evaluate it directly on an edge device, without relying on an intermediate simulation in the cloud. Pipelines can be converted into a demonstrable proof of concept in just a few minutes, rapidly reducing time to production deployment. It offers a more direct path to implementation at the edge for citizen developers, the company said.
The initial release of Palette Edgematic delivers a visual experience for developers to evaluate five optimized pipelines and more than 40 different models, covering object detection, tracking, classification, semantic segmentation and instance segmentation, among other use cases.
“We are setting a new standard in simplifying edge ML once and for all,” said SiMa founder and Chief Executive Krishna Rangasayee. “SiMa is laser-focused on removing any friction or obstacles for organizations looking to embed AI and ML into their products and services. If we do our job right, nobody doing edge ML should ever have to resort to embedded hand coding again.”
Andy Thurai, vice president and principal analyst at Constellation Research Inc., told SiliconANGLE there’s a major opportunity for running AI at the edge, as opposed to running it in cloud-based data centers, as most workloads are done now. However, he said the challenge is that edge AI locations may not have any consistent networking or internet connectivity, and they struggle with limitations in terms of power supply. Even the weather conditions at edge locations can hamper AI, he said.
“It is difficult to establish the systems, software and connectivity at the edge that are needed to do proper AI,” Thurai explained. “SiMa is trying to avoid this by providing the entire platform for machine learning development and deployment on a chip. Technically, one can create, build and deploy machine learning applications locally using this single chip.”
The analyst added that this could be suitable for AI applications in warehouses, factory floors, weather stations, autonomous cars, drones and other areas. “SiMa’s solutions might not help people train large language models like ChatGPT at the edge, but there are many AI applications that operate on much smaller models,” Thurai said. “So SiMa could be a good fit in these situations, especially given its performance numbers against Nvidia’s GPUs.”
SiMa said Palette Edgematic is being offered as a free visual extension for the Palette software development platform, as part of its ambition to scale AI and machine learning at the edge. According to SiMa’s senior vice president of engineering and operations Gopal Hegde, most machine learning applications are deployed in the cloud. Very few models have been deployed at the edge so far because of a lack of tools that simplify the process.
“Customers have to hand optimize models and the entire end to end application to get necessary performance and accuracy,” Hegde said. “It typically takes multiple weeks to evaluate and several months to deploy edge ML applications.”
Cambrian-AI Research analyst Karl Freund said machine learning at the edge offers transformative possibilities. “What SiMa has introduced with Palette Edgematic will rapidly fuel innovation by giving any organization the power to see what’s possible at the edge through the testing and optimizing their applications, and significantly accelerating their time to deployment,” he said.
THANK YOU