Join our daily and weekly newsletters for the latest updates and exclusive content on industry-leading AI coverage. Learn More
Edge computer chip and software startup SiMa.ai, fresh off a $70 million funding round from industry heavyweights including Dell Technologies Capital, is expanding its foothold in the edge AI market with the release of a new, smaller, lower power chip: MLSoC Modalix.
At 6 nanometers, it comes in way smaller than the San Jose, California-based startup’s prior MLSoC chip of 16 nanometers.
Building on the company’s ONE Platform for Edge AI, this new offering is designed to support advanced AI models such as Convolutional Neural Networks (CNNs), Transformers, and Generative AI, all while delivering industry-leading energy efficiency and scalable performance.
In a video call interview, SiMa.ai CEO Krishna Rangasayee shared the excitement surrounding the launch. “We’re extending the momentum we have as being the one platform for AI and introducing a capability not only in silicon but also the software that goes along with it,” he told VentureBeat.
Where will the Modalix family end up? Rangasayee says it’s perfect for “industrial automation, healthcare, smart vision systems, aerospace and defense, and anywhere there’s a need for multimodal elements.”
As Rangasayee pointed out, “Robotics, embodied AI, and sensory information are the future. Modalix is perfect for that. It’s about generative AI-centric architecture driving new applications, like human-robot interactions.”
But, the ultimate vision is even more ambitious. Rangasayee says the chip could help usher in an age where “every appliance, every device, is going to be capable of human-like capacity. So it’ll be able to talk, express, and visualize.”
Pushing the boundaries of edge AI
Generative AI is rapidly transforming industries, but has largely been confined so far to desktop PCs and mobile devices. Now, thanks to Sima.AI and its competition — namely GPU leader Nvidia, which also offers edge chips in its Orin and Xavier families — the technology is advancing to allow for powerful AI models to be deployed in dedicated, specialized devices out in the field and the factory floor, such as robotic arms and drones.
“We are consistently in real-life applications 10x better than an immediate competitor, and now this further extends where it will be more than 10x of anybody else,” Rangasayee claimed.
SiMa.ai’s MLSoC Modalix platform is designed to handle multimodal AI processing, integrating inputs such as text, images, and audio. It can run variants of Meta’s Llama 2-7B parameter model right on it, a huge potential unlock for reasoning at the edge.
As Rangasayee noted, “People are combining reality. So you could get audio with video with text, and the input could be any of these, and the output could be a combination. That’s the second big shift we’re addressing.”
The MLSoC Modalix family introduces several configurations ranging from 25 to 200 TOPS, each engineered to handle demanding AI workloads while minimizing power consumption.
According to Rangasayee, this new platform represents a leap in capability: “Modalix bridges the evolutions that have happened in the last two years. Now you can run everything from CNNs to the latest cutting-edge models on a single chip.”
SiMa.ai’s technology, designed specifically for edge applications, addresses key challenges in the field, including performance-per-watt.
“One key technical merit is frames per second per watt, or inferences per second per watt. In real-life applications, we’re 10x better than our immediate competitors, and Modalix extends that lead,” Rangasayee explained.
The goal, he added, is for customers to no longer worry about power and cooling constraints. “With Modalix, it’s a checkbox: low power, high performance – reshaping what’s possible at the edge.”
Endorsed by industry players
The potential of SiMa.ai’s MLSoC Modalix family has not gone unnoticed by industry leaders. Arye Barnehama, CEO of Elementary, expressed enthusiasm for the platform’s energy efficiency and high performance, which aligns with Elementary’s vision inspection systems.
Similarly, Vaibhav Ghadiok, CTO of Hayden AI, highlighted the platform’s ability to enable multimodal AI on power-constrained edge devices.
The MLSoC Modalix family also benefits from SiMa.ai’s Palette Edgematic software stack, a no-code, drag-and-drop platform designed to make AI deployment accessible to non-specialist developers.
Incorporating innovations such as an integrated Image Signal Processor (ISP), PCIe Gen 5 support, and eight Arm Cortex-A65 CPUs, the platform is built to handle a range of AI workloads. SiMa.ai’s approach allows for seamless integration of AI into existing workflows.
A growing edge AI market
SiMa.ai’s latest product launch also signals its ambition to compete with industry giants like Nvidia. While Nvidia dominates in cloud-based AI applications, SiMa.ai is focusing on a niche where real-time, on-device processing is critical. Last year, Rangasayee emphasized that SiMa.ai’s chips outperformed Nvidia’s in terms of both performance and power efficiency for edge AI applications.
As edge AI continues to grow, analysts predict the global edge computing market will double in the coming years, driven by advances in AI and increased demand for real-time decision-making at the edge.
SiMa.ai’s MLSoC Modalix family is well-positioned to meet this demand, offering a platform capable of processing multimodal AI models on a single chip.
With the launch of the MLSoC Modalix family, SiMa.ai is extending its leadership in edge AI. The platform’s high performance, energy efficiency, and ease of deployment make it a compelling option for industries seeking to harness the power of AI at the edge.
With strong backing from investors like Dell Technologies Capital and a growing list of industry partners, SiMa.ai is poised to lead the next wave of AI innovation at the edge.
Source link
