SiMa.ai uses agent developer tools to reduce physical AI deployment from months to days.

Artificial intelligence chip startup SiMa Technologies Inc. today launched Palette Neat, a purpose-built agent AI development environment for creating applications that connect the physical world to AI models.

This new tool can be easily combined with the company’s Modalix MLSoC System-on-Module or the company’s new PCIe companion card, allowing developers to quickly build apps that can see, learn from, adapt to, and interact with the real world. This is a vision of physical AI, where artificial intelligence models will be used to power robotics, self-driving cars, drones, industrial machinery, aerospace platforms, smart vision, and more.

“SiMa.ai is an AI software company that builds its own silicon,” said Founder and CEO Krishna Rangasay. “Today, we are delivering the industry’s first agent development environment for physical AI.”

Rangasayee explained that with Palette Neat, today’s developers can capture knowledge and simply speak or input their ideas directly to an AI agent, which then builds abstract thinking directly into low-level computational code. This eliminates the months of effort traditionally spent wiring, porting, and integrating applications into new silicon.

Often, whenever a company upgrades or offers a new form factor, developers spend months or even weeks taking new ideas and mapping them to new silicon. Neat is designed to eliminate this pain point by allowing developers to build entire systems by writing what they need with natural language commands, freeing engineers to focus on system-level differentiation and sorting out nuances.

The agent environment handles the heavy lifting and reduces development cycle time by mapping application code directly to silicon. SiMa says this allows them to reuse application code without having to rewire everything, saving 90% of legacy software investments from other platforms.

Giving developers an alternative to Nvidia in edge AI

The company’s Modalix SoM can run multiple large-scale language models simultaneously alongside vision and sensor models, all at less than 10 watts. This serves as a drop-in replacement for similar Nvidia Corp. system-on-module form factors, giving it the ability to deliver edge AI to physical AI deployments in situations that do not require board redesign.

Rangasayee claims that using Neat and Modalix has the potential to free developers from the market pressure that graphics processing units have on the physical AI inference development market as they scale physical AI. Most developers learn to build for the Nvidia ecosystem because cloud-based GPU hardware is CUDA-based and Nvidia already owns nearly 39% of edge AI, with Qualcomm Inc. in second place at about 20%.

Developers tend to gravitate toward tools they already know and what they have at hand. By providing an agent platform that allows you to migrate to a new system without having to sit down and learn something completely new, you can prototype and experiment without spending years or months training to see the benefits.

“Until now, developers have not had a seamless, performance-optimized alternative to physical AI,” Rangasayee said. Therefore, Modalix was designed “pin-to-pin” to fit into Nvidia’s Orin SoM. “Rather than retrofitting power-hungry data center GPUs for the edge, we deliver energy-efficient performance in stark contrast to what they already have.”

SiMa anticipates that developers will seek the opportunity to run their software and models on a variety of hardware that can manage different opportunities such as scalability, performance per watt, thermal envelope, and dynamic demand. At a time when the market is largely dominated by Nvidia, the company hopes to carve out a slice of its customer base by offering a different option than incumbents and give them the tools to make the transition.

Image: SiliconANGLE/Microsoft Designer